hckrnws

Reading these comments aren't we missing the obvious?

Claude Code is a lock in, where Anthropic takes all the value.

If the frontend and API are decoupled, they are one benchmark away from losing half their users.

Some other motivations: they want to capture the value. Even if it's unprofitable they can expect it to become vastly profitable as inference cost drops, efficiency improves, competitors die out etc. Or worst case build the dominant brand then reduce the quotas.

Then there's brand - when people talk about OpenCode they will occasionally specify "OpenCode (with Claude)" but frequently won't.

Then platform - at any point they can push any other service.

Look at the Apple comparison. Yes, the hardware and software are tuned and tested together. The analogy here is training the specific harness,caching the system prompt, switching models, etc.

But Apple also gets to charge Google $billions for being the default search engine. They get to sell apps. They get to sell cloud storage, and even somehow a TV. That's all super profitable.

At some point Claude Code will become an ecosystem with preferred cloud and database vendors, observability, code review agents, etc.

> At some point Claude Code will become an ecosystem with preferred cloud and database vendors, observability, code review agents, etc.

i've been wondering how anthropic is going to survive long term. If they could build out an infrastructure and services to complete with the hyperscalers but surfaced as a tool for claude to use then maybe. You pay Anthropic $20/user/month for ClaudeCode but also $100k/month to run your applications.

Anthropic is going to be on the losing side with this. Models are too fungible, it's really about vibes, and Claude Code is far too fat and opinionated. Ironically, they're holding back innovation, and it's burning the loyalty the model team is earning.

the fat and opinionated has always been true for them (especially compared to openai), and to all appearances remains a feature rather than a bug. i can’t say the approach makes my heart sing, personally, but it absolutely has augured tremendous success among thought workers / the intelligensia

I thought Anthropic would fall after OpenAI, but they just might be racing to the bottom faster here.

Maybe for coding but the number of normie users flooding to Claude over OAI is huge.

I think you have it exactly backwards, and that "owning the stack" is going to be important. Yes the harness is important, yes the model is important, but developing the harness and model together is going to pay huge dividends.

https://mariozechner.at/posts/2025-11-30-pi-coding-agent/

This coding agent is minimal, and it completely changed how I used models and Claude's cli now feels like extremely slow bloat.

I'd not be surprised if you're right in that this is companies / management will prefer to "pay for a complete package" approach for a long while, but power-users should not care for the model providers.

I have like 100 lines of code to get me a tmux controls & semaphore_wait extension in the pi harness. That gave me a better orchestration scheme a month ago when I adopted it, than Claude has right now.

As far as I can tell, the more you try to train your model on your harness, the worse they get. Bitter lesson #2932.

> I'd not be surprised if you're right in that this is companies / management will prefer to "pay for a complete package" approach for a long while

I mean I suspect for corporate usage Microsoft already has this wrapped up with Microsoft & GitHub Co-Pilots.

OpenAI, Anthropic, Google, Microsoft certainly desire path dependence but the very nature of LLMs and intelligence itself might make that hard unless they can develop models which truly are differentiated (and better) from the rest. The Chinese open source models catching up make me suspect that won't happen. The models will just be a commodity. There is a countdown clock for when we can get Opus 4.6+ level models and its measured in months.

The reason these LLM tools being good is they can "just do stuff." Anthropic bans third party subscription auth? I'll just have my other tool use Claude Code in tmux. If third party agents can be banned from doing stuff (some advanced always on spyware or whatever), then a large chunk of the promise of AI is dead.

Amp just announced today they are dumping IDE integration. Models seem to run better on bare-bones software like Pi, and you can add or remove stuff on the fly because the whole things open source. The software writes itself. Is Microsoft just trying to cram a whole new paradigm in to an old package? Kind of like a computer printer. It will be a big business, but it isn't the future.

At scale, the end provider ultimately has to serve the inference -- they need the hardware, data centers & the electricity to power those data centers. Someone like Microsoft can also provide a SLA and price such appropriately. I'll avoid a $200/month customer acquisition cost rant, but one user, running a bunch of sub agents, can spend a ton of money. If you don't own a business or funding source, the way state of the art LLMs are being used today is totally uneconomical (easy $200+ an hour at API prices.)

36+ months out, if they overbuild the data centers and the revenue doesn't come in like OpenAI & Anthropic are forecasting, there will be a glut of hardware. If that's the case I'd expect local model usage will scale up too and it will get more difficult for enterprise providers.

(Nothing is certain but some things have become a bit more obvious than they were 6 months ago.)

Thinking about this a little more -> "nature of LLMs and intelligence"

Bloated apps are a material disadvantage. If I'm in a competitive industry that slow down alone can mean failure. The only thing Claude Code has going for it now is the loss making $200 month subsidy. Is there any conceivable GUI overlay that Anthropic or OpenAI can add to make their software better than the current terminal apps? Sure, for certain edge cases, but then why isn't the user building those themselves? 24 months ago we could have said that's too hard, but that isn't the case in 2026.

Microsoft added all of this stuff in to Windows, and it's a 5 alarm fire. Stuff that used to be usable is a mess and really slow. Running linux with Claude Code, Codex, or Pi is clearly superior to having a Windows device with neither (if it wasn't possible to run these in Windows; just a hypothetical.)

From the business/enterprise perspective - there is no single most important thing, but having an environment that is reliable and predictable is high up there. Monday morning, an the Anthropic API endpoint is down, uh oh! In the longer term, businesses will really want to control both the model and the software that interfaces with it.

If the end game is just the same as talking to the Star Trek computer, and competitors are narrowing gaps rather than widening them (e.g. Anthropic and OpenAI releases models minutes from each other now, Chinese frontier models getting closer in capability not further), then it is really hard to see how either company achieves a vertical lock down.

We could actually move down the stack, and then the real problem for OpenAI and Anthropic is nVidia. 2030, the data center expansion is bust, nVidia starts selling all of these cards to consumers directly and has a huge financial incentive to make sure the performant local models exist. Everyone in the semiconductor supply chain below nvidia only cares about keeping sales going, so it stops with them.

Maybe nvidia is the real winner?

Also is it just me or does it now feel like hn comments are just talking to a future LLM?

That was true more mid last year, but now we have a fairly standard flow and set of core tools, as well as better general tool calling support. The reality is that in most cases harnesses with fewer tools and smaller system prompts outperform.

The advances in the Claude Code harness have been more around workflow automation rather than capability improvements, and truthfully workflows are very user-dependent, so an opinionated harness is only ever going to be "right" for a narrow segment of users, and it's going to annoy a lot of others. This is happening now, but the sub subsidy washes out a lot of the discontent.

If Claude Code is so much better why not make users pay to use it instead of forcing it on subscribers?

You're right, because owning the stack means better options for making tons of money. Owning the stack is demonstrably not required for good agents, there are several excellent (frankly way better than ol' Claude Code) harnesses in the wild (which is in part why so many people are so annoyed by Anthropic about this move - being forced back onto their shitty cli tool).

The competition angle is interesting - we're already seeing models like Step-3.5-Flash advertise compatibility with Claude Code's harness as a feature. If Anthropic's restrictions push developers toward more open alternatives, they might inadvertently accelerate competitor adoption. The real question is whether the subscription model economics can sustain the development costs long-term while competitors offer more flexible terms.

I don't think many are confused about why Anthropic wants to do this. The crux is that they appear to be making these changes solely for their own benefit at the expense of their users and people are upset.

There are parallels to the silly Metaverse hype wave from a few years ago. At the time I saw a surprising number of people defending the investment saying it was important for Facebook to control their own platform. Well sure it's beneficial for Facebook to control a platform, but that benefit is purely for the company and if anything it would harm current and future users. Unsurprisingly, the pitch to please think of this giant corporation's needs wasn't a compelling pitch in the end.

"Training the specific harness" is marginal -- it's obvious if you've used anything else. pi with Claude is as good as (even better! given the obvious care to context management in pi) as Claude Code with Claude.

This whole game is a bizarre battle.

In the future, many companies will have slightly different secret RL sauces. I'd want to use Gemini for documentation, Claude for design, Codex for planning, yada yada ... there will be no generalist take-all model, I just don't believe RL scaling works like that.

I'm not convinced that a single company can own the best performing model in all categories, I'm not even sure the economics make it feasible.

Good for us, of course.

> pi with Claude is as good as (even better! given the obvious care to context management in pi) as Claude Code with Claude

And that’s out of the box. With how comically extensible pi is and how much control it gives you over every aspect of the pipeline, as soon as you start building extensions for your own, personal workflow, Claude Code legimitely feels like a trash app in comparison.

I don’t care what Anthropic does - I’ll keep using pi. If they think they need to ban me for that, then, oh well. I’ll just continue to keep using pi. Just no longer with Claude models.

>Claude Code is a lock in, where Anthropic takes all the value.

I wouldn't all the value, but how else are you going to run the business? Allow other to take all the value you provide?

> Reading these comments aren't we missing the obvious?

AI companies: "You think you own that code?"

???

Use an API Key and there's no problem.

They literally put that in plain words in the ToS.

Using an API key is orders of magnitude more expensive. That's the difference here. The Claude Code subscriptions are being heavily subsidized by Anthropic, which is why people want to use their subscriptions in everything else.

They are subsidized by people who underuse their subscriptions. There must be a lot of them.

I think the people who use more than they pay for vastly outnumber those who pay for more than they use. It takes intention to sign up (not the default, like health care) and once you do, you quickly get in the habit of using it.

Of course, like gym memberships. VC’s subsidizing powerlifters…

This move feels poorly timed. Their latest ad campaigns about not having ads, and the goodwill they'd earned lately in my book was just decimated by this. I'm sure I'm not the only one who's still just dipping their toes into the AI pool. And am very much a user that under utilizes what I pay for because of that. I have several clients who are scrambling to get on board with cowork. Eliminating API usage for subscription members right before a potentially large wave of turnover not only chills that motivation it signals a lack of faith in their marketing, which from my POV, put out the only AI super bowl campaign to escape virtually unscathed.

> the goodwill they'd earned lately in my book was just decimated by this

That sounds absurd to me. Committing to not building in advertising is very important and fundamental to me. Asking people who pay for a personal subscription rather than paying by the API call to use that subscription themselves sounds to me like it is. Just clarifying the social compact that was already implied.

I WANT to be able to pay a subscription price. Rather like the way I pay for my internet connectivity with a fixed monthly bill. If I had to pay per packet transmitted, I would have to stop and think about it every time I decided to download a large file or watch a movie. Sure, someone with extremely heavy usage might not be able to use a normal consumer internet subscription; but it works fine for my personal use. I like having the option for my AI usage to operate the same way.

The problem with fixed subscriptions in this model is that the service has an actual consumption cost. For something like internet service, the cost is primarily maintenance, unless the infrastructure is being expanded. But using LLMs is more like using water, where the more you use it, the greater the depletion of a resource (electricity in this case, which is likely being produced with fossil fuel which has to be sourced and transported, etc). Anthropic et al would be setting themselves up for a fall if they allow wholesale use at a fixed price.

There is a lot more vc cash.

There are. It's like healthcare, the healthy don't use it as much and pay for the sick.

Be the economics as they may, there is no lock in as OP claims.

This statement is plainly wrong.

If you boost and praise AI usage, you have to face the real cost.

Can't have your cake and eat it, too.

The people mad about this feel they are entitled to the heavily subsidized usage in any context they want, not in the context explicitly allowed by the subsidizer.

It's kind of like a new restaurant started handing out coupons for "90% off", wanting to attract diners to the restaurant, customers started coming in and ordering bulk meals then immediately packaging them in tupperware containers and taking it home (violating the spirit of the arrangement, even if not the letter of the arrangement), so the restaurant changed the terms on the discount to say "limited to in-store consumption only, not eligible for take-home meals", and instead of still being grateful that they're getting food for 90% off, the cheapskate customers are getting angry that they're no longer allowed to exploit the massive subsidy however they want.

It might be some confirmation bias here on my part but it feels as if companies are becoming more and more hostile to their API users. Recently Spotify basically nuked their API with zero urgency to fix it, redit has a whole convoluted npm package your obliged to use to create a bot, Facebook requires you to provide registered company and tax details even for development with some permissions. Am I just old man screaming at cloud about APIs used to being actually useful and intuitive?

They put no limits on the API usage, as long as you pay.

Here, they put limits on the "under-cover" use of the subscription. If they can provide a relatively cheap subscription against the direct API use, this is because they can control the stuff end-to-end, the application running on your system (Claude Code, Claude Desktop) and their systems.

As you subscribe to these plans, this is the "contract", you can use only through their tools. If you want full freedom, use the API, with a per token pricing.

For me, this is fair.

> If they can provide a relatively cheap subscription against the direct API use

Except they can't. Their costs are not magically lower when you use claude code vs when you use a third-party client.

> For me, this is fair.

This is, plain and simple, a tie-in sale of claude code. I am particularly amused by people accepting it as "fair" because in Brazil this is an illegal practice.

> This is, plain and simple, a tie-in sale of claude code. I am particularly amused by people accepting it as "fair" because in Brazil this is an illegal practice

I am very curious what is particularly illegal about this. On the sales page nowhere do they actually talk about the API https://claude.com/pricing

Now we all know obviously the API is being used because that is how things work, but you are not actually paying a subscription for the API. You are paying for access to Claude Code.

Is it also illegal that if you pay for Playstation Plus that you can't play those games on an Xbox?

Is it illegal that you can't use third party netflix apps?

I really don't want to defend and AI company here but this is perfectly normal. In no other situation would we expect access to the API, the only reason this is considered different is because they also have a different service that gives access to the API. But that is irrelevant.

It's basically the difference between pro-market capitalism and pro-business capitalism. The value to the society comes from competition in the market and from the businesses' ability to choose freely how they do business. When those two goals are in conflict, which one should be prioritized?

Anthropic provides an API third-party clients can use. The pro-market position is that the API must be available at every pricing tier, as the benefits from increased competition outweigh the imposed restrictions to business practices. The pro-business position is that Anthropic must be allowed to choose which tiers can use the API, as the benefits from increased freedom outweigh the reduced competition in the market.

Like I mentioned somewhere else I can see why some people think they are entitled to do this and I also fully understand wanting to do it from a cost standpoint.

While I do personally disagree with thinking that you should be able to do this when it was never sold in that way, at the end of the day as a customer you can choose if you want to use the product in the way that they are saying or use something else if you don’t want to support that model.

However the person I was responding too brought up legality which is a very different discussion.

Imagine if video service came with a free TV that watched you, and was really opinionated about what you watch, and you could only watch your videos on the creeper TV.

Then I would not use it because it does not work the way I want it to work...

But if that is the service they are making and they are clear about what it is when you sign up... That does not make it illegal.

I can see why people think they should be entitled to do this, but it does not align with how they are selling the service or how many other companies sell services. In most situations you don't get unlimited access to the individual components of how a service works (the API), you are expected to use the service (in this case Claude Code) directly.

> That does not make it illegal.

"Both parties are okay with the terms" is far from being sufficient to make something "legal".

Tie-in sales between software and services is not different from price dumping. If any of the Big Tech corporations were from any country that is not the US, the FTC would be doing anything in their power to stop them.

> Tie-in sales between software and services is not different from price dumping.

I disagree, in many cases what you are specifically paying for is the combination of the software and the service that are designed to work together. And in many cases do not work independent of eachother.

There are countless cases of this, that what you are paying for is a thing that is made up of a piece of software and a serverside component. MMO's (and gaming in general) being a major example of this, but so are many of the apps I pay for subscriptions for on my phone.

The actual technical implementation of how it works is irrelevant when it is clear what it is you are paying for.

> "Both parties are okay with the terms" is far from being sufficient to make something "legal".

True but the opposite is also true, just because you don't like the terms it does not make it illegal.

> in many cases what you are specifically paying for is the combination of the software and the service that are designed to work together

And in many cases like Claude Code and the Anthropic models, they can and do work perfectly independently.

> True but the opposite is also true, just because you don't like the terms it does not make it illegal.

This is not me "not liking it". Like I said somewhere else in this thread: these types of tie-in are illegal in Brazil. This practice is clearly not done to favor the consumer. You can bet that if the US was anything closer to a functional democracy and the laws were not written by lobbyists, this would be illegal in the US as well.

What law is actually being broken in Brazil?

Are MMO’s illegal in Brazil? Is PlayStation Plus illegal in Brazil? Is Spotify, Apple Music, etc etc etc also illegal in Brazil?

It would be ridiculous to argue that I could pay for a subscription to World of Warcraft and make my own third party client to play the game with. (Obviously you are free to argue it all you want but I would be very surprised if this was actually illegal).

> And in many cases like Claude Code and the Anthropic models, they can and do work perfectly independently.

Unless I am mistaken Claude Code does not have a local model built into it, so it requires a server side component to work?

As far as the Anthropic models, yes like many other services they ALSO have a public API that is separate from the subscription that you are paying for.

The critical difference here being that in the subscription it is very clear that you are paying for “Claude Code” which is a combination of an application and a server side component. It makes no claims about API usage as part of your subscription, again the technical implementation of the service you are actually paying for “Claude Code” is irrelevant.

When it comes to “Claude Code” for all that we should care about, again given that “Claude Code” is what you are paying for, they could be sending the information to Gemini or or a human looks at it. Because it’s irrelevant to the end user when it comes to the technical implementation since you are not being granted access to any other parts of the system directly.

> What law is actually being broken in Brazil?

"Tie-in sale": the business practice where a seller conditions the sale of one product (the tying good) on the buyer’s agreement to purchase a different product (the tied good).

The examples you are giving are not "tie-in" sales because the service from Playstation Plus, Spotify, Apple Music, etc is the distribution of digital goods.

> Unless I am mistaken Claude Code does not have a local model built into it, so it requires a server side component to work?

Which part are you not understanding?

I don't care about Claude Code. I do not want it and do not need it. All I care about is the access to the models through the client that I was already using!

> When it comes to “Claude Code” for all that we should care about, again given that “Claude Code” is what you are paying for.

No, it is not! I paid for Claude Pro. Claude != Claude Code.

> "Tie-in sale": the business practice where a seller conditions the sale of one product (the tying good) on the buyer’s agreement to purchase a different product (the tied good)

I will keep my response to this part in particular limited because I have limited understanding of this law. However based on doing a little bit of searching around the law is not as cut and dry as you are presenting it to be. It is possible that Claude code would fall under being fine under that law or no one has gone after them. I honestly don’t know and I don’t feel like having an argument that it is highly likely both of us don’t fully understand the law.

That being said I do question how exactly “Claude code” differs from those services as a digital good.

> I don't care about Claude Code. I do not want it and do not need it. All I care about is the access to the models through the client that I was already using!

OK! That is not what you’re paying for as part of Claude Pro, end of story. You are not paying for the API. It is no different that the people that have a free plan and can only chat through the web and the app also don’t get access to the API even though it is obviously using an API to access those endpoints as well.

Or are you also going to argue that free users should have access to the API because they are already using them in the browser.

> No, it is not! I paid for Claude Pro. Claude != Claude Code.

Claude Code is one of the features you are paying for as part of Claude Pro so yes in a way you are paying for it. And again not on that list is the API.

Claude Pro = claude.ai, and they made no changes to that arrangement. Both claude.ai and Claude Pro are products built on top of the Claude API. You are free to buy access to the Claude API itself, with or without the other two, but the pricing is different because the price of claude.ai and Claude Code includes the API charges they incur.

> but the pricing is different because the price of claude.ai and Claude Code includes the API charges they incur.

If that was true, then getting equivalent usage of the API without claude.ai and Claude Code should cost less, not more.

You can try to find all sorts of explanations for it, at the end of the day is quite simple: they are subsidizing one product in order to grow the market share, and they are doing it at a loss now, because they believe they will make up for it later. I understand the reasoning from a business point of view, but this doesn't mean they are entitled to their profits. I do not understand people that think we simply accept their premise and assume they can screw us over just because they asked and put it on a piece of paper.

> All I care about is the access to the models through the client that I was already using!

But that's not a product that they're offering. That ability was an undesired (from their business perspective) trait that they're now rectifying.

> But that's not a product that they're offering

Of course it was.

  - It was possible to do it.
  - OpenCode did not break any security protocol in order to integrate with them. 
  - OAuth is *precisely* a system to let third-party applications use their resources.

> (from their business perspective)

So what?!

Basically, they set up an strategy they thought it was going to work in their favor (offer a subsidized service to try to lock in customers), someone else found a way to turn things around and you believe that we should be okay with this?!

Honestly, I do not understand why so many people here think it is fine to let these huge corporations run the same exploitation playbook over and over again. Basically they set up a mouse trap full of cheese and now that the mice found a way to enjoy the cheese without getting their necks broken, they are crying about it?

> Of course it was.

You'd have to point me to an authoritative source on that (explicitly saying users are allowed to use their models via private APIs in apps of the user's choosing). If something isn't explicitly provided in the contract, then it can be changed at any point in any way without notice.

Honestly, I'm not big on capitalism in general, but I don't understand why people should expect companies to provide things exactly the way they want at exactly the prices they would like to be charged (if at all). That's just not how the world/system works, or should, especially given there are so many alternatives available. If one doesn't like what's happening with some service, then let the wallet do the talking and move to another. Emigration is a far more effective message than complaining.

> Tie-in sales between software and services

Look at it this way: the service that you're accessing is really a (primarily desired) side-effect of the software. So re subscriptions, what they're actually providing are the apps (web, desktop, etc), and the apps use their service to aid the fulfillment of their functionality. Those wanting direct access to the internal service can get an API key for that purpose. That's just how their product offering is structured.

Video service does work like that. They call it DRM.

The Telly comes with a second screen for ads that you're not allowed to shut off. https://www.telly.com/

That’s definitely a pitch lol

I think you just described American cable boxes... Except they charge us a monthly fee and an additional monthly fee for the box.

Or any smart tv with free ip tv.

Is that not most if not all smart TVs today? Basically nearly every TV made and sold right now?

I've heard they actually cache the full Claude Code system prompt on their servers and this saves them a lot of money. Maybe they cache the MCP tools you use and other things. If another harness like Opencode changes that prompt or adds significantly to it, that could increase costs for them.

What I don't understand is why start this game of cat and mouse? Just look at Youtube and YT-DLP. YT-DLP, and the dozens of apps that use it, basically use Youtube's unofficial web API and it still works even after Youtube constantly patches their end. Though now, YT-DLP has to use a makeshift JS interpreter and maybe even spawn Chromium down the line.

Some people drop out of the game as it gets harder. I've basically stopped looking at youtube videos unless I want it enough to download it (and wait if the current workarounds broke) with how much they've clamped down on no-account usage. Most I suspect just give in to the company's terms.

I just use NewPipe all the way.

[dead]

Unless it's illegal in more places, I think they won't care. In my experience, the percentage of free riders in Brazil is higher (due to circumstances, better said).

But they could charge the third-party client for access to the API.

While the cost may not be lower the price certainly can be if they are operating like any normal company and adding margin.

> Except they can't. Their costs are not magically lower when you use claude code vs when you use a third-party client.

I don't have a dog in this fight but is this actually true? If you're using Claude Code they can know that whatever client-side model selection they put into it is active. So if they can get away with routing 80% of the requests to Haiku and only route to Opus for the requests that really need it, that does give them a cost model where they can rely on lower costs than if a third-party client just routes to Opus for everything. Even if they aren't doing that sort of thing now, it would be understandable if they wanted to.

It (CC) does have a /models command, you can still decide to route everything to Opus if you just want to burn tokens I guess it's not default so most wouldn't, but still, people willing to go to a third party client are more likely that kind of power user anyway

They still have the total consumption under their control (*bar prompt caching and other specific optimizations) where in the past they even had different quotas per model, it shouldn't cost them more money, just be a worse/different service I guess

> it shouldn't cost them more money

As things are currently, better models mean bigger models that take more storage+RAM+CPU, or just spend more time processing a request. All this translates to higher costs, and may be mitigated by particular configs triggered by knowledge that a given client, providing particular guarantees, is on the other side.

That’s kind of the point. Even if users can choose which model to use (and apparently the default is the largest one), they could still say (For roughly the same cost): your Opus quota is X, your Haiku quota is Y, go ham. We’ll throttle you when you hit the limit.

But they don't want the subscription to be quota'd like that. The API automatically does that though, as different models use different amounts of tokens when generating responses, and the billing is per token. And quite literally is having the user account for the actual costs of usage, which is the thing said users are trying to avoid, on their own terms, and getting upset about when they aren't.

> It (CC) does have a /models command, you can still decide to route everything to Opus if you just want to burn tokens I guess it's not default so most wouldn't

Opus is claude code's default model as of sometime recently (around Opus 4.6?)

That’s not how Claude Code works. It’s not like a web chatbot with a layer that routes based on complexity of request.

You don't control what happens when a request hits their endpoint though.

> Their costs are not magically lower when you use claude code vs when you use a third-party client.

If subsidizing that offering is a good hook to get higher paying API users on board, then some of that cost is a customer aquisition cost, whereas the cost to them of providing the API doesn't have the same proportion that they can justify as a customer acquisition cost.

I absolutely have zero concerns about their cost to acquire new customers. As a (former) customer, all I am concerned is the freedom to consume the service I am paying for however I see fit.

Is there any service in the world that gives you complete freedom on how you consume it? I can’t think of one.

Netflix: limits number of devices and stream quality and offline use.

AWS: does not allow any number of applications (spamming, crypto mining, adult content)

Airlines: do not allow smoking, boom boxes

Is there any service that gives complete freedom?

I am not asking for "complete" freedom, am I? Can you please argue in good faith and not resort to cheap rhetoric tricks?

I think what most people don't realize is running an agent 24/7 fully automated is burning a huge hole in their profitability. Who even knows how big it is. It could be getting it on the 8/9 figures a day for all we know.

There's this pervasive idea left over from the pre-llm days that compute is free. You want to rent your own H200x8 to run your Claude model, that's literally going to cost $24/hour. People are just not thinking like that. I have my home PC, it does this stuff I can run it 24/7 for free.

there are usage limits preventing you from running it 24/7 on all subscriptions tiers

This bolster's OP's point, to an extent

I understand you mean for free in the sense that you don't pay a third party to use it, however let's no forget that you still use the power grid and that's not free. Also worth to note that energy prices have increased worldwide.

Depending on utilisation and good use of low-power or sleep (or full off) states when things aren't actively processing, it can still be a _lot_ cheaper to run things at home than on a rented service. Power costs have increased a lot in recent years, but so have compute-per-watt ratios and you are not paying the that indirect compute price when the processors are asleep or off whereas with subscription access to LLMs you are paying at least the base subscription each month even if you don't use it at all in that period. Much the same as the choice between self-hosting an open-source project or paying for a hosted instance - and in both cases people don't tend to consider the admin cost (for some of us the admin is “play time”!) so the self-hosted option it does practically feel free.

Maintaining hardware also isn't free. Time is money.

Time is money if you have another good use of that time. If you like spending that time doing something, then it's literally free.

Sure it's $24/hour, but it'll crank through tens of thousands of tokens per second --- those beefy GPUs are meant for large amounts of parallel workflow. You'll never _get_ that many tokens for a single request. That's why the mathematics work when you get dozens or hundreds of people using it.

No. The sauce is in KV caching: when to evict, when to keep, how to pre-empt an active agent loop vs someone who are showing signs of inactivity at their pc, etc.

Coder doing the coding should use subscription, and now they ban the choice of your preferred ide for agentc coding. API is for automation not coding. I'm going to cancel their subscription today, I already use codex with opencode.

This is honestly the key difference here. I’m morally okay with using Claude Max Whatever with something like OpenCode because it’s literally the same thing from the usage pattern perspective. Plugging Nanoclaw into it is a whole another thing.

It probably doesn't help that the creator of OpenClaw just got hired by Anthropic's competitor.

This sounds like engineering, finance, and legal got together and decided they were in an untenable position if OpenAI started nudging OpenClaw to burn even more tokens on Anthropic (or just never optimize) + continually updated workarounds to using subscription auth. But I'm sure OpenAI would never do something like that...

At the end of the day, it's the same 'fixed price plan for variable use on a constrained resource' cellular problem: profitability becomes directly linked to actual average usage.

> OpenAI started nudging OpenClaw to burn even more tokens on Anthropic

Not possible: OpenClaw is run by a foundation, and is open source, which means OpenAI has no leverage to do such a thing.

I don't see how it's fair. If I'm paying for usage, and I'm using it, why should Anthropic have a say on which client I use?

I pay them $100 a month and now for some reason I can't use OpenCode? Fuck that.

You aren't paying for usage, you are paying for the product that the subscription is offered to. If you are paying for usage, well, that's their billed by token-usage API plan, which they are quite happy for you to use with any client you want.

Even worse, if I'm paying a subscription for the product, and I don't want to use the product, what's it to them?

You are free to not use it. You are not free to use the API provided specifically for the product, which you are not explicitly paying for, for a different product.

You can of course use OpenCode or any other project with the API, which is also offered as a separate product. People just don't want to do that because it's not subsidized, ie. more expensive. But the entire reason it's subsidized is that Anthropic can use the data to improve their product.

> But the entire reason it's subsidized is that Anthropic can use the data to improve their product.

This is grade A, absolute crap. It's subsidized because everyone else is subsidizing it, and everyone is doing it because they are trying to lock their consumer share.

The solution is quite simple. Just get the FTC to forbid tie-in sales so that we don't get the huge corporations using their infinite resources to outlive the competition. Anthropic/Amazon/Google/OpenAI/Facebook can offer any type of service they want, but if the access to the API costs $X when offered standalone, then that is the baseline price for anything that depends on the API to work.

I'm fine with this as well. I just dislike everyone here presenting it like this is Anthropic being unreasonable. Given the product that is offered and why it's being offered, this is completely reasonable to do.

I don't use the Anthropic subscriptions either.

Do you not understand that they run the regular subscription at a huge loss? In exchange they require you to stick with Claude Code.

You are free to use the API.

Why can't they run the API at a huge loss too then??

Because if you do everything at a huge loss, you run out of runway very quickly.

I’m not saying this is a bubble; I don’t know whether than it is or not. But if it is a bubble, I’ve only seen one bubble this big in tech. When that bubble popped, the only companies that survived kept their costs in check. Some raised a round right before the bubble popped but cost control was always part or their survival.

To be honest the fact that they are already clamping down makes me think their situation is getting strained.

Stop giving money to the company that doesn't give you what you want.

It is important that the company knows why they are losing customers, though.

I canceled my Claude subscription (other reasons) and they had an "exit interview" question of why you canceled. They know why.

Internal sales data is probably a lot more effective and attended to than HN posts.

> If I'm paying for usage

You are not paying for usage. You are paying for usage via their application.

If their business plan is based on how quickly a human can enter requests and react to the results, and Claude Code is optimized for that, why should you be allowed to use an alternative client that e.g. always tries to saturate the token limits?

Btw API is not for coding, it's designed for pipelines, automation, products. They just kill competition making better software like opencode.

Is that your belief to what their API should be used for?

But a) I'm not doing that and b) they can just ban that, like they have rate limits. Why ban OpenCode?

They have rate limits, but they also want to control the nozzle, and not all their users use all their allocation all the time.

In reality, heavy subscription users are subsidized by light subscription users. The rate limits aren't everything.

If agent harnesses other than Claude Code consume more tokens than average, or rather, if users of agent harnesses other than CC consume more tokens than average, well, Anthropic wouldn't be unhappy if those consumers had to pay more for their tokens.

> If agent harnesses other than Claude Code consume more tokens than average, or rather, if users of agent harnesses other than CC consume more tokens than average

Do they, though?

The speculative reasoning I've seen is that they have optimizations in their CC client that reduces their costs. If that's true, I think it's fair that they can limit subscription usage to their client. If you don't want those optimizations and prefer more freedom, use the API.

They rather have yolo permissions to run arbitrary code on your machine and phone home all the time, then opencode having it and phoning home all the time.

You're touching on the eternal App Store debate. "It's my phone, I should be able to install whatever I want on it!" Which is true, but also hasn't been true since the mid-90s (early 2000s at the latest).

Not quite though. You can install Claude's apps wherever they're supported, and maybe even fiddle with the source code (I'm unsure). And you can use any other coding apps that you want. The only real restriction is how those apps are allowed to connect to the providers' services, which are running on their servers, etc. There's a movement from "my local domain" to "their remote domain", and they're allowed to have full control of theirs as you - would prefer, I think - full control of yours.

Now just imagine the rug-pull if you ever get really dependent on the product.

Read the ToS, you are paying to use their products. If you want to use other products that integrate with the Anthropic LLMs they offer a product which is the API. You can use Opencode by connecting your API and being charged per token.

Doesn't that make sense? If you use it more you get charged more, if you use it less you get charged less.

Sure, that's why I'm cancelling my max subscription because I'm tied to opencode :)

But you understand that they changed the ToS today, and that's what I'm complaining about, right? "Read the ToS" isn't an answer to "I don't like this ToS change".

I didn't see today's ToS change. But this was always against ToS. OpenClaw specifically is built against tech that breaks ToS.

Probably the ToS change was to make it more clear.

To be fair, the developer is the one breaking the ToS in the most significant way, breaking boilerplate reverse engineering clauses.

But the user also is very aware that they are doing something funny, in order to authenticate, the user is asked to authorize Claude Code, n ot Opencode or OpenClaw, it's clearly a hack and there is no authorization from Anthropic to OpenClaw, and you are not giving Anthropic authorization to give access to OC, the user asks Anthropic to give access to Claude Code, the only reason this works is because OC is pretending to be Claude Code.

The bottom line issue is that as a user you are paying for a subscription to a package that includes an expected usage. That package is not metered, but it is given on the condition that you will use it as it is expected to be used, by chatting manually with the chatbot, which results in a reasonable expected token usage. By using a program that programatically calls the chat interface, the token consumption increases beyond what was part of the original deal, and so the price should be different.

A similar scenario would be if you go to an all you can eat buffet, you pay for a single person, but then you actually unload an army of little clones that start eating the whole buffet. Technically it was an all you can eat buffet and you paid the price right? Well no, come on, don't play dumb.

Their subscriptions aren't cheap, and it has nothing really to do with them controlling the system.

It's just price differentiation - they know consumers are price sensitive, and that companies wanting to use their APIs to build products so they can slap AI on their portfolio and get access to AI-related investor money can be milked. On the consumer-facing front, they live off branding and if you're not using claude code, you might not associate the tool with Anthropic, which means losing publicity that drives API sales.

It would be less of an issue if Claude-Code was actually the best coding client, and would actually somehow reduce the amount of tokens used. But it's not. I get more things done with less tokens via OpenCode. And in the end, I hit 100% usage at the end of the week anyway.

The problem is incentives. The organization selling the per-token model doesn't have the incentive, at scale to have you reduce token consumption. Other technologies do, hence adding value.

I'm with the parent comment. It was inevitable Netflix would end password-sharing. It was inevitable you'd have to pick between freeform usage-based billing and a constrained subscription experience. Using the chatbot subscription as an API was a weird loophole. I don't feel betrayed.

They tier it. So you are limited until you pay more. So you can't just right away get the access you need.

It doesn't really make sense to me because the subscriptions have limits too.

But I agree they can impose whatever user hostile restrictions they want. They are not a monopoly. They compete in a very competitive market. So if they decide to raise prices in whatever shape or form then that's fine.

Arbitrary restrictions do play a role for my own purchasing decisions though. Flexibility is worth something.

[dead]

I don't and would never pay for an LLM, but presumably they also want for force ads down your throat eventually, yea? Hard to do if you're just selling API access.

But the idea of an API is more to encourage creativity and other people/companies building products and services on or around your systems. This used to be seen as a positive as it would mean you were an important cog in other peoples products and so more users, exposure, brand awareness etc.

Anthropic do of course sell API access and you can of course use any product with it.

1:29 until you're able to push to the `main` branch.

Please enjoy these messages from our sponsors.

Compared to the heavily subsidized subscriptions, I don't think API is sold at loss.

Also why would you create a throwaway for this question? Are you trying to rage bait?

> Also why would you create a throwaway for this question? Are you trying to rage bait?

You should never question anyone's route to privacy :)

> Are you trying to rage bait?

If you have to ask, it's probably not rage bait. I'm just too lazy to come up with a username.

Every garden eventually becomes a walled garden once enough people are inside.

#earthisamoat

[flagged]

What a weird metaphor for a company which most likely wants nothing more than growing markets.

It's not a metaphor at all, just racism.

[flagged]

Spotify are probably reacting to https://annas-archive.li/blog/backing-up-spotify.html where basically the whole archive was downloaded

that was later.

It was known publicly later.

See [1] for a solution.

[1] https://i.programmerhumor.io/2025/03/778c56a79115a582edb9949...

Can you sell ads via api? If answer is no then this “feature” would be at the bottom of the list

They can sell API access via transparent pricing.

Instead, many, many websites (especially in the music industry) have some sort of funky API that you can only get access to if you have enough online clout. Very few are transparent about what "enough clout" even means or how much it'd cost you, and there's like an entire industry of third-party API resellers that cost like 10x more than if you went straight to the source. But you can't, because you first have to fulfill some arbitrary criteria that you can't even know about ahead of time.

It's all very frustrating to deal with.

Of course they can [1].

Though, in this case, you get free API access to the model.

[1]: https://x.com/badlogicgames/status/2017063228094709771

Plus, use of the API is a way to avoid ads. So double-strike against good/available APIs.

There is a world where approaches like HTTP 402 are implemented to monetize API usage.

Please get this token signed by our ad partner to enable your next ten requests.

What kind of ads they sell in terminal Claude code? Are you bor?

You're correct in your observations. In the age of agents, the walls are going up. APIs are no longer a value-add; they're a liability. MCP and the equivalent will be the norm interface. IMO.

I think that these companies are understanding that as the barrier to entry to build a frontend gets lower and lower, APIs will become the real moat. If you move away from their UI they will lose ad revenue, viewer stats, in short the ability to optimize how to harness your full attention. It would be great to have some stats on hand and see if and how much active API user has increased decreased in the last two years, as I would not be surprised if it had increased at a much faster pace than in the past.

> the barrier to entry to build a frontend gets lower

My impression is the opposite: frontend/UI/UX is where the moat is growing because that's where users will (1) consume ads (2) orchestrate their agents.

I agree with you - we are saying the same thing, by restricting their API or making less developer friendly, they want you to be captive in their UI. This might not be true for Anthropic or OpenAI - another child commenter made a comment about ads in CLI, I would not be surprised if in a while we will have product placements in LLM responses exactly as we have it in movies - not a plain ad but just a slightly less subliminal suggestion.

I don't think we'll get ads in the enterprise version. There will be ads for individuals and usual business pricing for businesses.

I just think that OAI/Anthropic will try to keep both types of users locked into their walled garden via the UI.

The APIs may have a future, but at our own peril and zero guarantees. It's a tool to create traction and demonstrate capabilities to devs.

It’s objectively easier to build a frontend now and therefore that moat is disappearing. What you can argue is the moat is in incumbent advantage at the UI layer, not the UI itself.

What ad revenue? In their terminal cli?

What is given can be taken away. Despite the extra difficult this is why unofficial methods (e.g. scraping) are often superior. Soon we'll see more fully independent data scraping done by cameras and microphones.

„Open Access APIs are like a subway. You use them to capture a market and then you get out.“

— Erdogan, probably.

It's just the continued slow death of the open internet

I don't it's particularly hard to figure it out: APIs have been particularly at risk of being exploited for negative purposes due the explosion of AI powered bots

This trend well predates widespread use of chatbots.

I’m predicting that there would be a new movement to make everything an MCP. It’s now easier to consume an api by non technical people.

APIs leak profit and control vs their counterpart SDK/platforms. Service providers use them to bootstrap traffic/brand, but will always do everything they can to reduce their usage or sunset them entirely if possible.

Google now wants $30,000 a month for customsearch (minimum charge), up from 1c per search or thereabouts in January 2026...

Facebook doing that is actually good, to protect consumers from data abuse after incidents like cambridge analytica. They are holding businesses who touches your personal data responsible.

> Facebook doing that is actually good, to protect consumers from data abuse after incidents like cambridge analytica.

There is nothing here stopping cambridge analytica from doing this again, they will provide whatever details needed. But a small pre launch personal project work that might use a facebook publishing application can't be developed or tested without first going through all the bureaucracy.

Nevermind the non profit 'free' application you might want to create on the FB platform, lets say a share chrome extension "Post to my FB", for personal use, you can't do this because you can't create an application without a company and IVA/TAX documents. It's hostile imo.

Before, you could create an app, link your ToS, privacy policy etc, verify your domain via email, and then if users wanted to use your application they would agree, this is how a lot of companies still do it. I'm actually not sure why FB do this specifically.

Facebook knew very early and very well about the data harvesting that was going on at Cambridge Analytica through their APIs. They acted so incredibly slowly and not-harsh that it's IMO hard to believe that they did not implicitly support it.

> to protect consumers

We are talking about Meta. They have never, and will never, protect customers. All they protect is their wealth and their political power.

Is it? I’ve never touched Facebook api, but it sounds ridiculous that you need to provide tax details for DEVELOPMENT. Can’t they implement some kind of a sandbox with dummy data?

You can mock their API all you want for development and there are many pre-built options for that, but it you want to touch their systems, they're sending a very clear signal. You must be a corporate with an RBO. Seems prudent to me.

WhatsApp takes bot spam very very seriously, and as a result, there is zero bot spam.

Before you can sign up to build a WhatsApp bot, you need to jump through a million hoops, and after that, every automated message template must be vetted by Meta before it can be sent out, apple style.

I'm glad of this, because unlike SMS and other messaging platforms, WhatsApp is spam free and a pleasure to use.

> WhatsApp is spam free and a pleasure to use.

At least here in Italy whatsapp is a spam house unless you actively update the default privacy settings in-app. There is no discernable difference between SMS and WhatsApp to spammers.

WhatsApp is certainly not spam free. I get spam over WhatsApp, less than on telegram but still more than zero.

They just want people to use facebook. If you can see facebook content without being signed in they have a harder time tracking you and showing you ads.

Not sure how relevant this comment is

Given the Cambridge Analytica scandal, I don’t take too much issue to FB making their APIs a little tougher to use

Everyone has heard the word "enshittification" at this point and this falls in line. But if you haven't read the book [0] it's a great deep dive into the topical area.

But the real issue is that these companies, once they have any market leverage, do things in their best interest to protect the little bit of moat they've acquired.

[0] https://www.mcdbooks.com/books/enshittification

There is no moat except market saturation and gate keeping for most platforms.

APIs are the best when they let you move data out and build cool stuff on top. A lot of big platforms do not really want that anymore. They want the data to stay inside their silo so access gets slower harder and more locked down. So you are not just yelling at the cloud this feels pretty intentional.

You're not wrong. Reddit & Elon started it and everyone laughed at them and made a stink. But my guess is the "last dying gasp of the freeloader" /s wasn't enough to dissuade other companies from jumping on the bandwagon, cause fiduciary responsibility to shareholders reigns supreme at the end of the day.

But this ban is precisely on circumventing the API.

This is sort of true!

Spotify in particular is just patently the very worst. They released an amazing and delightful app sdk, allowing for making really neat apps in the desktop app in 2011. Then cancelled it by 2014. It feels like their entire ecosystem has only ever gone downhill. Their car device was cancelled nearly immediately. Every API just gets worse and worse. Remarkable to see a company have only ever such a downward slide. The Spotify Graveyard is, imo, a place of singnificantly less honor than the Google Graveyard. https://web.archive.org/web/20141104154131/https://gigaom.co...

But also, I feel like this broad repulsive trend is such an untenable position now that AI is here. Trying to make your app an isolated disconnected service is a suicide pact. Some companies will figure out how to defend their moat, but generally people are going to prefer apps that allow them to use the app as they want, increasingly, over time. And they are not going to be stopped even if you do try to control terms!

Were I a smart engaged company, I'd be trying to build WebMCP access as soon as possible. Adoption will be slow, this isn't happening fast, but people who can mix human + agent activity on your site are going to be delighted by the experience, and that you will spread!

WebMCP is better IMHO than conventional APIs because it layers into the experience you are already having. It's not a separate channel; it can build and use the session state of your browsing to do the things. That's a huge boon for users.

I really hope someone from any of those companies (if possible all of them) would publish a very clear statement regarding the following question: If I build a commercial app that allows my users to connect using their OAuth token coming from their ChatGPT/Claude etc. account, do they allow me (and their users) to do this or not?

I totally understand that I should not reuse my own account to provide services to others, as direct API usage is the obvious choice here, but this is a different case.

I am currently developing something that would be the perfect fit for this OAuth based flow and I find it quite frustrating that in most cases I cannot find a clear answer to this question. I don't even know who I would be supposed to contact to get an answer or discuss this as an independent dev.

EDIT: Some answers to my comment have pointed out that the ToS of Anthropic were clear, I'm not saying they aren't if taken in a vacuum, yet in practice even after this being published some confusion remained online, in particular regarding wether OAuth token usage was still ok with the Agent SDK for personal usage. If it happens to be, that would lead to other questions I personally cannot find a clear answer to, hence my original statement. Also, I am very interested about the stance of other companies on this subject.

Maybe I am being overly cautious here but I want to be clear that this is just my personal opinion and me trying to understand what exactly is allowed or not. This is not some business or legal advice.

I don't see how they can get more clear about this, considering they have repeatedly answered it the exact same way.

Subscriptions are for first-party products (claude.com, mobile and desktop apps, Claude Code, editor extensions, Cowork).

Everything else must use API billing.

The biggest reason why this is confusing is the Claude Agent SDK[0] will use subscription/oauth credentials if present. The terms update implies that there's some use cases where that's ok and other use cases (commercial?) where using their SDK on a user's device violates terms.

[0] https://platform.claude.com/docs/en/agent-sdk/overview

Had the same question, comment below quotes their docs saying Agent SDK using oAuth token is also not allowed.

It doesn't quote their docs, it quotes the page this post is about.

The SDK is Claude Code in a harnesss, so it works with your credentials the same way CC does.

But they're stating you can only use your subscription for your personal usage, not someone else's for their usage in your product.

I honestly think they're being short sighted not just giving a "3rd party quota" since they already show users like 4 quotas.

If the fear is 3rd party agents screwing up the math, just make it low enough for entry level usage. I suspect 3rd party token usage is bi-modal where some users just need enough to kick tires, but others are min-maxing for how mamy tokens they can burn as if that's its own reward

How can they be clearer that the Agents SDK is not allowed?

> OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.

I agree, it'd actually be great if they did give maybe $5 or $10 worth of API tokens per month to max subscribers, since they're likely to be the most likely to actually build stuff that uses the Claude APIs.

I built a quick thing to download YouTube videos and transcribe them using with whisper, but it kind of feels clunky to summarize them using the claude CLI, even though that works.

just ran into this myself. I got Claude Code to build a tool that calls Claude for <stuff>. Now I have to create a console account and do the API thing and it sucks balls.

And at that point, you might as well use OpenRouter's PKCE and give users the option to use other models..

These kinds of business decisions show how these $200.00 subscriptions for their slot/infinite jest machines basically light that $200.00 on fire, and in general how unsustainable these business models are.

Can't wait for it all to fail, they'll eventually try to get as many people to pay per token as possible, while somehow getting people to use their verbose antigentic tools that are able to inflate revenue through inefficient context/ouput shenanigans.

I think the subscription pricing exists because it’s a far more palatable way to bill people for day to day personal use.

I used Claude back when API per token pricing was the only option and it was bad for all the usual reasons pay-per-use sucks compared to flat billing: you’re constantly thinking about cost. Like trying to watch a Netflix video with a ticker in the corner counting up the cents you owe them.

I don’t understand your claim that they want people paying per token - the subscription is the opposite of that, and it also has upsides for them as a business since most people don’t saturate the usage limits, and the business gets to stuff a bunch of value-adds on a bundle offering which is generally a more lucrative and enticing consumer pricing model.

The bundle only works if it’s +EV for them. A lot of analyses (though not all - it’s complicated) say that the $200/mo bundle (and certainly the $20/mo bundle) costs more than that for most users, and the bundle is currently a loss leader. If so, then eventually prices will need to go up, and API per usage pricing will seem much more attractive.

Even if it more expensive, people will prefer subscription pricing over pay per use.

When you ask it to do something and it goes off the rails, the payment plans have wildly different effects:

Subscription- oh well, let's try again with a different prompt

Pay per use- I just wasted money, this product sucks

Even if it is less common than not, it has an outsized impact on how people feel using it.

And it also caps your maximum expenses. A subscription user don’t have to worry something goes wrong and end up with a huge bill.

At least theoretically, the bill would work on a “wallet” system, where you fill up your account with $X every month, and then you’re charged per use. That keeps there from being a huge bill, worst case you’re just on hold until the next fill up.

I'm not going to say what platform but it's an agentic coding tool, I know for a fact the platform loses in the mid $200.00s on a $20.00 plan. 10:1 loss leader for customer acquisition is crazy, and they'll have to make that up in the future somehow, they're all fumbling on how to vendor lock their customers, and its not necessarily clear they're going to be able to.

I expect some big falls from 10 figure businesses in the next year or two as they realize this is impossible. They've built an industry on the backs of gambling addicts and dopamine feins (I'm generalizing but this is a thing with LLM users (just read vibe coders posts on twitter, they're slot machine users). Ask sports betting operators from back in 2019-2022 how it worked out for them when they tried to give out 1-2k a year to attract new customers, and then realized their customers will switch platforms in an instant they see a new shiny offer. Look up the Fanduel Founders "exit" for an insight into this.

They have to eventually stop catering to the slot machine users, which are generally paying for these hugely lossy flat rate subscriptions, and somehow get them used to a different type of payment model, or cater strictly to enterprise... Which also aren't going to tolerate paying 20k a month in tokens per developer, is my guess.... Lots of delicate pricing problems to figure out for all these companies.

That's crazy. I'm already barely willing to pay $10/month on Github Copilot. A product I love. Best value for money.

If they pump it up to $200 (or to $20). I'll simply use crappier local model. It won't be as good. But I already own my gaming PC that can run local models, and electricity is cheap.

Calling LLM users "slot machine users" makes no sense and tells me that you just have an axe to grind.

True, they should be called "slop machine users" instead.

The cost difference is pretty staggering for the same usage. Being on the sub hacks your reward system to push you to be productive, legitimately hitting limits feels like a win, and you start looking for ways to max your utilization %. A lot of people get quite obsessive about it. The sub is 100% the innovation that makes Claude Code "work."

If the pay-per-use cost predictable enough, it’s less of an issue. That’s how electricity works and it’s fine.

The issue with Claude Code is it’s not at all obvious how any given task or query translates to cost. I was finding some days I spent very little and other days cost a fortune despite what seemed to me to be similar levels of usage.

People 100% want subscriptions in this space.

The alternative is AWS where you need to be a billing expert to keep costs locked at $20/month.

It’s been obvious from the start that the $200 point is the free tier

You are talking about Anthropic and indeed compared to OpenAI or GitHub Copilot they have seemed to be the ones with what I would personally describe as a more restrictive approach.

On the other hand OpenAI and GitHub Copilot have, as far as I know, explicitly allowed their users to connect to at least some third party tools and use their quotas from there, notably to OpenCode.

What is unclear to me is whether they are considering also allowing commercial apps to do that. For instance if I publish a subscription based app and my users pay for the app itself rather than for LLM inference, would that be allowed?

Same question here. A while ago I read rumors OpenAI might build a "Login with OpenAI" (comparable to login with Apple, Facebook, Google) so people can also use their existing sub in commercial apps. Hope it's true.

Then why does the SDK support subscription usage? Can I at least use my subscription for my own use of the SDK?

What if you wrap the service using their Agent SDK?

That should be fine, because it's still using their tooling. And this seems like the better way to go. I have a couple of tools that work like this. I think the issue is mostly 3rd party harnesses that seek to do the same as Claude Code. And it seems reasonable that Anthropic decides how you can use the subscription, because it's heavily subsidized. Get a Claude $200 sub and max out the usage limits, then compare that usage to the cost of using their API. The difference is significant, which is why people are getting multiple $200 subs rather than paying for API usage (and I have seen reports where they are cracking down on this as well.)

It literally says in the linked page it's not fine.

Okay, I was mistaken. The tooling I was speaking of uses Claude Code rather than the SDK. One uses the Zed ACP protocol. I'm not sure about the other. I should have said Claude Code rather than the SDK. For example, I can run a session through one of the tools, and then access that session directly in Claude Code. It's still Claude though. It seems the important element is that you're not using OAuth tokens from a sub to use in a different tool. If you go through Claude Code, then Claude Code is handling everything and giving your tool the output. Thanks for the correction.

Yeah, I'm designing some stuff right now and I'm having it run Claude Code headless rather than use the Agents SDK for this exact reason.

Quick question but what if I use claude code itself for the purpose?

https://github.com/rivet-dev/sandbox-agent/tree/main/gigacod... [I saw this inShow HN: Gigacode – Use OpenCode's UI with Claude Code/Codex/Amp] (https://news.ycombinator.com/item?id=46912682)

This can make Opencode work with Claude code and the added benefit of this is that Opencode has a Typescript SDK to automate and the back of this is still running claude code so technically should work even with the new TOS?

So in the case of the OP. Maybe Opencode TS SDK <-> claude code (using this tool or any other like this) <-> It uses the oauth sign in option of Claude code users?

Also, zed can use the ACP protocol itself as well to make claude code work iirc. So is using zed with CC still allowed?

> I don't see how they can get more clear about this, considering they have repeatedly answered it the exact same way.

This is confusing quite frankly, there's also the claude agent sdk thing which firloop and others talked about too. Some say its allowed or not. Its all confusing quite frankly.

That’s very clearly a no, I don’t understand why so many people think this is unclear.

You can’t use Claude OAuth tokens for anything. Any solution that exists worked because it pretended/spoofed to be Claude Code. Same for Gemini (Gemini CLI, Antigravity)

Codex is the only one that got official blessing to be used in OpenClaw and OpenCode, and even that was against the ToS before they changed their stance on it.

Is Codex ok with any other third party applications, or just those?

Yes. You can build third party applications on top of codex app server. All open source. https://developers.openai.com/codex/app-server/

  Codex app-server is the interface Codex uses to power rich clients (for example, the Codex VS Code extension). Use it when you want a deep integration inside your own product.

I think it's permissible. Zed uses it to power their Codex integration. OpenAI has been quite vocal about it.

By default, assume no. The lack of any official integration guide should be a clear sign. Even saying that you reverse-engineer Codex for apps to pretend to be Codex makes it clear that this is not an officially endorsed thing to do

Codex is Open Source though, so I wonder at what stage me adding features to Codex is different from me starting a new project and using the subscription.

But I believe OpenAI does let you use their subscription in third parties, so not an issue anyway.

Interested to know this too

But why does it matter which program consumes the tokens?

Why does it matter to the free buffet manager where do you consume the food? We may never know.

Because it could be over longer time periods than buffet hours.

They must be getting something out of it, because we sure aren't.

Cory Doctorow has a word for this..

They think their position is strong enough to lock users in. I'm not so sure.

It's enshittification - for those who didn't know.

They'll own entire pipeline interface, conduit, backend. Interface is what people get habitual to. If I am a regular user of Claude Code, I may not shift to competitor for 10-20% gains in cost.

They want that sweet vendor lock-in.

Presumably because their flat rate pricing is based off their ability to manage token use via their first-party tools.

A third-party tool may be less efficient in saving costs (I have heard many of them don't hit Anthropic LLMs' caches as well).

Would you be willing to pay more for your plan, to subsidize the use of third-party tools by others?

---

Note, afaik, Anthropic hasn't come out and said this is the reason, but it fits.

Or, it could also just be that the LLM companies view their agent tools as the real moat, since the models themselves aren't.

But wouldn't a less efficient tool simply consume your 5-hour/weekly quota faster? There's gotta be something else, probably telemetry, maybe hoping people switch to API without fighting, or simply vendor lock-in.

Presumably most people also do not use their full quota when using the official client, whereas third-party clients could be set up to start back up every 5 hours to use 100% of the quota every day and week.

It's the whole "unlimited storage" discussion again.

> But wouldn't a less efficient tool simply consume your 5-hour/weekly quota faster?

Maybe.

First, Anthropic is also trying to manage user satisfaction as well as costs. If OpenCode or whatever burns through your limits faster, are you likely to place the blame on OpenCode?

Maybe a good analogy was when DoorDash/GrubHub/Uber Eats/etc signed up restaurants to their system without their permission. When things didn't go well, the customers complained about the restaurants, even though it wasn't their fault, because they chose not to support delivery at scale.

Second, flat-rate pricing, unlike API pricing, is the same for cached vs uncached iirc, so even if total token limits are the same, less caching means higher costs.

> are you likely to place the blame on OpenCode?

am I? Probably, but I get your point that your average user would blame Anthropic instead.

> even if total token limits are the same, less caching means higher costs

Not really, flat-rate pricing simply gives you a fixed token allotment, so less caching means you consume your 5-hour/weekly allotment faster.

What if I'm only willing to pay if it support by tool of choice? Would you pay for a streaming service that enforces a certain TV brand?

Given the latest changes on Claude Code where they hide the actions

https://news.ycombinator.com/item?id=47033622

it's likely more the other way around. They control how fast your subscription tokens are burned

> What if I'm only willing to pay if it support by tool of choice?

I don’t want to say that you won’t be missed but they will get over it.

It is pretty obviously no. API keys billed by the token, yes, Oauth to the flat rate plans no.

> OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.

If you look at this tweet [1] and in particular responses under it, it still seems to me like some parts of it need additional clarification. For instance, I have seen some people interpret the tweet as meaning using the OAuth token is actually ok for personal experimentation with the Agent SDK, which can be seen as a slight contradiction with what you quoted. A parent tweet also mentioned the docs clean up causing some confusion.

None of this is legal advice, I'm just trying to understand what exactly is allowed or not.

[1] https://x.com/trq212/status/2024212380142752025?s=10

Read the actual ToS. What you describe is NOT allowed.

That tweet is from a product leader on Claude Code itself...

A tweet is not a ToS.

Then they should speak to legal about fixing the ToS before making public statements about their intentions with it. It won't look good to show up at arbitration and have to explain why your public comms contradict your ToS.

What flatrate?

Pro and Max are both limited

Flat rate does not imply unlimited.

What else would it mean?

That you are charged a single fixed fee regardless of usage.

Nothing about that prevents a usage cap.

That you are buying a bundle and it doesnt matter how much of the bundle you use you pay the same amount every billing period?

So if I buy entry to the swimming hall that allows me to be there for 4 hours but also allows me to leave earlier you would call that a flat rate?

I have never noticed there are people who interpret it that way.

If it allows you 4 hours total per month yea

>A flat fee, also referred to as a flat rate or a linear rate refers to a pricing structure that charges a single fixed fee for a service, regardless of usage.

That's one definition. There are others.

I'm sure you can use context clues to figure this one out. You're so close! Just put the pieces together.

I think you're just trying to see ambiguity where it doesn't exist because the looser interpretation is beneficial to you. It totally makes sense why you'd want that outcome and I'm not faulting you for it. It's just that, from a POV of someone without stake in the game, the answer seems quite clear.

> OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai.

I think this is pretty clear - No.

So it’s forbidden to use the Claude Mac app. I would say the ToS as it is, can’t be enforced

Anthropic has published a very clear statement. It's "no".

Does https://happy.engineering/ need to use the API keys or can use oauth? It's basically a frontend for claude-cli.

It doesn't even touch auth right?

""" Usage policy

Acceptable use Claude Code usage is subject to the Anthropic Usage Policy. Advertised usage limits for Pro and Max plans assume ordinary, individual usage of Claude Code and the Agent SDK """

That tool clearly falls under ordinary individual use of Claude code. https://yepanywhere.com/ is another such tool. Perfectly ordinary individual usage.

https://yepanywhere.com/sdk-auth-clarification.html

The TOS are confusing because just below that section it talks about authentication/credential use. If an app starts reading api keys / credentials, that starts falling into territory where they want a hard line no.

If it's a wrapper that invokes the `claude` binary then I believe it's fine.

Is there a way to legally or even practically prevent this? `claude` CLI execution in a shell is certainly included in the subscription - it’s the product.

Practically; yes. MMOs have been doing this kind of thing (Preventing alteration / automation of the client) for ages now.

Usually, it is already stated in their documentation (auth section). If a statement is vague, treat it as a no. It is not worth the risk when they can ban you at any time. For example, ChatGPT allows it, but Claude and Gemini do not.

https://developers.openai.com/codex/auth

Maybe I am missing something from the docs of your link, but I unfortunately don't think it actually states anything regarding allowing users to connect and use their Codex quota in third party apps.

https://x.com/thdxr/status/2013010664776683644

I can't find anything official from OpenAI, but they have worked with the OpenCode people to support using your ChatGPT subscription in OpenCode.

From TFA: “OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.”

The comment you are responding to is about ChatGPT/Codex, not Claude.

They're not asking if Claude forbids it. They're asking if OpenAI (Codex, specifically) allows it.

One set of applications to build with subscription is to use the claude-go binary directly. Humanlayer/Codelayer projects on GitHub do this. Granted those are not ideal for building a subscription based business to use oathu tokens from Claude and OpenaAI. But you can build a business by building a development env and gating other features behind paywall or just offering enterprise service for certain features like vertical AI(redpanada) offerings knowledge workers, voice based interaction(there was a YC startup here the other day doing this I think), structured outputs and workflows. There is lots to build on.

Not allowed. They've already banned people for this.

[dead]

I'm only waiting for OpenAI to provide an equivalet ~100 USD subscription to entirely ditch Claude.

Opus has gone down the hill continously in the last week (and before you start flooding with replies, I've been testing opus/codex in parallel for the last week, I've plenty of examples of Claude going off track, then apologising, then saying "now it's all fixed!" and then only fixing part of it, when codex nailed at the first shot).

I can accept specific model limits, not an up/down in terms of reliability. And don't even let me get started on how bad Claude client has become. Others are finally catching up and gpt-5.3-codex is definitely better than opus-4.6

Everyone else (Codex CLI, Copilot CLI etc...) is going opensource, they are going closed. Others (OpenAI, Copilot etc...) explicitly allow using OpenCode, they explicitly forbid it.

This hostile behaviour is just the last drop.

OpenAI forces users to verify with their ID + face scan when using Codex 5.3 if any of your conversations was redeemed as high risk.

It seems like they currently have a lot of false positives: https://github.com/openai/codex/issues?q=High%20risk

They haven't asked me yet (my subscription is from work with a business/team plan). Probably my conversations as too boring

I’m unsure exactly in what way you believe it has gone “down the hill” so this isn’t aimed at you specifically but more a general pattern I see.

That pattern is people complaining that a particular model has degraded in quality of its responses over time or that it has been “nerfed” etc.

Although the models may evolve, and the tools calling them may change, I suspect a huge amount of this is simply confirmation bias.

> Opus has gone down the hill continously in the last week

Is a week the whole attention timespan of the late 2020s?

We’re still in the mid-late 2020s. Once we really get to the late 2020s, attention spans won’t be long enough to even finish reading your comment. People will be speaking (not typing) to LLMs and getting distracted mid-sentence.

Reminded https://www.baen.com/Chapters/9781618249203/9781618249203___...

thanks. i really enjoyed it

Seems we're already there.

My brain trailed off after "won’t be long enough to even finish"...

That's still impressive, given your claim of being a fish...

The most impressive thing is that this looks like it is your only comment on this network ^^

ps: imafish may only be a fan of <https://mumband.bandcamp.com/track/if-i-were-a-fish>

I would even call it mid 2020s. I think in a couple years people's attention spans will be so short they won't even finish reading comments.

Unfortunately, and “Attention Is All You Need”.

oh shit we're in the late 2020's now

Sorry, I don’t agree. And I won’t be taking questions at this time.

The rate limit for my $20 OpenAI / Codex account feels 10x larger than the $20 claude account.

YES. I hit the rate limit in about ~15 mins on Claude. But it will take me a few hours with Codex. A/B testing them on the same tasks. Same $20/mo.

Opus 4.6 genuinely seems worse than 4.5 was in Q4 2025 for me. I know everyone always says this and anecdote != data but this is the first time I've really felt it with a new model to the point where I still reach for the old one.

I'll give GPT 5.3 codex a real try I think

Huh… I’ve seen this comment a lot in this thread but I’ve really been impressed with both Anthropic’s latest models and latest tooling (plugins like /frontend-design mean it actually designs real front ends instead of the vibe coded purple gradient look). And I see it doing more planning and making fewer mistakes than before. I have to do far less oversight and debugging broken code these days.

But if people really like Codex better, maybe I’ll try it. I’ve been trying not to pay for 2 subscriptions at once but it might be worth a test.

> And I see it doing more planning and making fewer mistakes than before

Anecdotally, maybe this is the reason? It does seem to spend a lot more time “thinking” before giving what feels like equivalent results, most of the time.

Probably eats into the gambling-style adrenaline cycles.

I asked Codex 5.3 and Opus 4.6 to write me a macos application with a certain set of requirements.

Opus 4.6 wrote me a working macos application.

Codex wrote me a html + css mockup of a macos application that didn't even look like a macos application at all.

Opus 4.5 was fine, but I feel that 4.6 is more often on the money on its implementations than 4.5 was. It is just slower.

I asked both to help me with a hardware bug. Codex kept trying things, being sure of what the problem is every time, and every time making it worse.

Opus went off and browsed my dependencies for ten minutes, and came back and solved the problem firs try.

Funnily enough I've been using Codex 5.3 on maximum thinking for bug hunting and code reviews and it's been really good at it (it's just seem to have a completely different focus than Opus.)

I generally don't like the way codex approaches coding itself so I just feed its review comments back in to Claude Code and off we go.

I just created an OpenCode skill where both these models will talk to each other and discuss bug-finding approaches.

In my experience, two different models together works much better than one, that's why this subscription banning is distressing. I won't be able to use a tool that can use both models.

Heh, I find Codex to be a far, far smarter model than Claude Code.

And there's a good reason the most "famous" vibe coders, including the OpenClaw creator all moved to Codex, it's just better.

Claude writes a lot more code to do anything, tons of redundent code, repeated code etc. Codex is only model I've seen which occasionally removes more code than it writes.

Codex has written me 3 very nice mac os applications in the past week lol

Literally a skill issue.

I agree with you. Codex 5.3 is good it's just a bit slower.

It is (slower), especially at xhigh setting. But if I have to redo things three times, keep confirming trivial stuff (Claude Code seems to keep changing the commands it uses to read code... once it uses "bash-read", once it uses "tree", once it uses "head" and I have to keep confirming permission), I definitely waste more time than give a command to codex (or in my case OpenCode + codex model) and come back after 10 minutes.

I was underwhelmed by Opus4.6. I didn’t get a sense of significant improvement, but the token usage was excessive to the point that I dropped the subscription for codex. I am suspect that all the models are so glib that they can create a quagmire for themselves in a project. I have not yet found a satisfying strategy for non-destructive resets when the systems own comments and notes poisons new output. Fortunately, deleting and starting over is cheap.

> And don't even let me get started on how bad Claude client has become

The latest versions of claude code have been freezing and then crashing while waiting on long running commands. It's pretty frustrating.

> I’m only waiting for OpenAI to provide an equivalet ~100 USD subscription to entirely ditch Claude.

I have a feeling Anthropic might be in for an extremely rude awakening when that happens, and I don’t think it’s a matter of “if” anymore.

No offense, but this is the most predicable outcome ever. The software industry at large does this over and over again and somehow we're surprised. Provide thing for free or for cheap, and then slowly draw back availability once you have dominant market share or find yourself needing money (ahem).

The providers want to control what AI does to make money or dominate an industry so they don't have to make their money back right away. This was inevitable, I do not understand why we trust these companies, ever.

because it's easier than paying $50k for local llm setup that might not last 5 years.

Well, yes. They know what they are doing. They know when given the option the consumer makes the affordable choice. I just don't have to like or condone their practices. Maybe instead of taking on billions of dollars of debt they should have thought about a business model that makes sense first? Maybe the collective "we" (consumers and investors, but especially investors) should keep it in our pants until the product is proven and sustainable?

It will be real interesting if the haters are right and this technology is not the breakthrough the investors assume it to be AFTER it is already sewn into everyone's work flows. Everyone keeps talking about how jobs will be displaced, yet few are asking what happens when a dependency is swept out from underneath the industry as a whole if/when this massive gamble doesn't pay off.

Whatever. I am squawking into the void as we just repeat history.

Or the companies can be transparent about their product roadmap. I can guarantee this enshittification was on the roadmap way before we knew about it. They let us operate under false information, that's just weak behavior.

[dead]

No offense taken here :)

First, we are not talking about a cheap service here. We are talking about a monthly subscription which costs 100 USD or 200 USD per month, depending on which plan you choose.

Second, it's like selling me a pizza and pretending I only eat it while sitting at your table. I want to eat the pizza at home. I'm not getting 2-3 more pizzas, I'm still getting the same pizza others are getting.

It's the most overrated model there is. I do Elixir development primarily and the model sucks balls in comparison to Gemini and GPT-5x. But the Claude fanboys will swear by it and will attack you if you ever say even something remotely negative about their "god sent" model. It fails miserably even in basic chat and research contexts and constantly goes off track. I wired it up to fire up some tasks. It kept hallucinating and swearing it did when it didn't even attempt to. It was so unreliable I had to revert to Gemini.

It might simply be that it was not trained enough in Elixir RL environments compared to Gemini and gpt. I use it for both ts and python and it's certainly better than Gemini. For Codex, it depends on the task.

My favorite conspiracy explanation:

Claude has gotten a lot of popular media attention in the last few weeks, and the influx of users is constraining compute/memory on an already compute heavy model. So you get all the suspected "tricks" like quantization, shorter thinking, KV cache optimizations.

It feels like the same thing that happened to Gemini 3, and what you can even feel throughout the day (the models seem smartest at 12am).

Dario in his interview with dwarkesh last week also lamented the same refrain that other lab leaders have: compute is constrained and there are big tradeoffs in how you allocate it. It feels safe to reason then that they will use any trick they can to free up compute.

all this because of a single week?

No, it's not the first time their models degrade for some time.

No developer writes the same prompt twice. How can you be sure something has changed?

I regularly run the same prompts twice and through different models. Particularly, when making changes to agent metadata like agent files or skills.

At least weekly I run a set of prompts to compare codex/claude against each other. This is quite easy the prompt sessions are just text files that are saved.

The problem is doing it enough for statistical significance and judging the output as better or not.

I suspect you may not be writing code regularly... If I have to ask Claude the same things three times and it keeps saying "You are right, now I've implemented it!" and the code is still missing 1 out of 3 things or worse, then I can definitely say the model has become worse (since this wasn't happening before).

What model were you using where this wasn't happening before?

I haven't experiences this with gpt-5.3-codex (xhigh) for example. Opus/Sonnet usually work well when just released, then they degrade quite regularly. I know the prompts are not the same every day or even across the day, but if the type of problems are always the same (at least in my case) and a model starts doing stupid things, then it means something is wrong. Everyone I know who uses Claude regularly, usually have the same esperience whenever I notice they degrade.

> I suspect you may not be writing code regularly...

You have no reason to suspect this.

When I use Claude daily (both professionally and personally with a Max subscription), there are things that it does differently between 4.5 and 4.6. It's hard to point to any single conversation, but in aggregate I'm finding that certain tasks don't go as smoothly as they used to. In my view, Opus 4.6 is a lot better at long standing conversations (which has value), but does worse with critical details within smaller conversations.

A few things I've noticed:

* 4.6 doesn't look at certain files that it use to

* 4.6 tends to jump into writing code before it's fully understood the problem (annoying but promptable)

* 4.6 is less likely to do research, write to artifacts, or make external tool calls unless you specifically ask it to

* 4.6 is much more likely to ask annoying (blocking) questions that it can reasonably figure out on it's own

* 4.6 is much more likely to miss a critical detail in a planning document after being explicitly told to plan for that detail

* 4.6 needs to more proactively write its memories to file within a conversation to avoid going off track

* 4.6 is a lot worse about demonstrating critical details. I'm so tired of it explaining something conceptually without it thinking about how it implements details.

Just hit a situation where 4.6 is driving me crazy.

I'm working through a refactor and I explicitly told it to use a block (as in Ruby Blocks) and it completely overlooked that. Totally missed it as something I asked it to do.

Ralph Wiggum would like a word

Same prompt assumes same context state. But I think you get what I mean.

The economic tension here is pretty clear: flat-rate subscriptions are loss leaders designed to hook developers into the ecosystem. Once third parties can piggyback on that flat rate, you get arbitrage - someone builds a wrapper that burns through $200/month worth of inference for $20/month of subscription cost, and Anthropic eats the difference.

What is interesting is that OpenAI and GitHub seem to be taking the opposite approach with Copilot/OpenCode, essentially treating third-party tool access as a feature that increases subscription stickiness. Different bets on whether the LTV of a retained subscriber outweighs the marginal inference cost.

Would not be surprised if this converges eventually. Either Anthropic opens up once their margins improve, or OpenAI tightens once they realize the arbitrage is too expensive at scale.

these subscriptions have limits.. how could someone use $200 worth on $20/month.. is that not the issue with the limits they set on a $20 plan, and couldn't a claude code user use that same $200 worth on $20/month? (and how do i do this?)

The limits in the max subscriptions are more generous and power users are generating loss.

I'm rather certain, though cannot prove it, that buying the same tokens would cost at least 10x more if bought from API. Anecdotally, my cursor team usage was getting to around 700$ / month. After switching to claude code max, I have so far only once hit the 3h limit window on the 100$ sub.

What Im thinking is that Anthropic is making loss with users who use it a lot, but there are a lot of users who pay for max, but don't actually use it.

With the recent improvements and increase of popularity in projects like OpenClaw, the number of users that are generating loss has probably massively increased.

I've spent $17.64 on on-demand usage in cursor with an estimated API cost of $350, mostly using Claude Opus 4.5. Some of this is skewed since subagents use a cheaper model, but even with subagents, the costs are 10x off the public API costs. Either the enterprise on-demand usage gets subsidized, API costs are 10x higher, or cursor is only billing their 10% surplus to cover their costs of indexing and such.

edit: My $40/month subscription used $662 worth of API credits.

Cursor also significantly upcharges compared to API pricing. Last I checked they were charging ~3X API prices for Anthropic models

More than 20x actually. According to ccusage I’ve consumed the equivalent of $4500 worth of API tokens in the last 30 days on my $200 subscription.

This exactly. I think this is why Anthropic simply don’t want 3rd party businesses to max out the subscription plans by sharing them across their own clients.

I'd agree on this. I ended up picking up a Claude Pro sub and am very less than impressed at the volume allowance. I generally get about a dozen queries (including simple follow up/refinements/corrections) across a relatively small codebase, with prompts structured to minimize the parts of the code touched - and moving onto fresh contexts fairly rapidly, before getting cut off for their ~5 hour window. Doing that ~twice a day ends up getting cut off on the weekly limit with about a day or two left on it.

I don't entirely mind, and am just considering it an even better work:life balance, but if this is $200 worth of queries, then all I can say is LOL.

Bumping into those limits is trivial, those 5 hour windows are anxiety inducing, and I guess the idea is to have a credit card on tap to pay for overages but…

I’m messing around on document production, I can’t imagine being on a crunch facing a deadline or dealing with a production issue and 1) seeing some random fuck-up eat my budget with no take backs (‘sure thing, I’ll make a custom docx editor to open that…’), 2) having to explain to my boss why Thursday cost $500 more than expected because of some library mismatch, or 3) trying to decide whether we’re gonna spend or wait while stressing some major issue (the LLM got us in it, so we kinda need the LLM to get us out).

That’s a lot of extra shizz on top of already tricky situations.

The usage limit on your $20/month subscription is not $20 of API tokens (if it was, why subscribe?). Its much much higher, and you can hit the equivalent of $20 of API usage in a few days.

I think you've stated this in reverse.

API limits are infinite but you'd blow through $20 of usage in a maybe 1 hours or less of intense Opus use.

The subscription at $20/mo (or $200) allows for vastly more queries than $20 would buy you via API but you are constrained by hourly/weekly limits.

The $20/mo sub user will take a lot longer to complete a high token count task (due to start/stop) BUT they will cap their costs.

So I’m not allowed to use the $20 plan and max out its limits?

Max out on their terms, not yours.

Their bet is that most people will not fill up 100% of their weekly usage for 4 consecutive weeks of their monthly plan, because they are humans and the limits impede long running tasks during working hours.

You can max it out via first party clients only.

I dont like it either, but its not an unreasonable restriction.

I do believe it's unreasonable. The limits are the limits, you reach them there's no more free lunch after.

Fix the limits, so the limits are reached at a rate that sustains their business.. ? obviously this WILL happen eventually when they need to pay for things.

"A rate that sustains their business" at the moment probably looks like API pricing or maybe even higher. That means subscriptions get significantly more expensive and/or limited, which is maybe where things are headed.

The median subscriber generates about 50% gross margin, but some subscribers use 10x the amount of inference compute as other subscribers (due to using it more...), and it's a positive skewness distribution.

I don't think it's a secret that AI companies are losing a ton of money on subscription plans. Hence the stricter rate limits, new $200+ plans, push towards advertising etc. The real money is in per-token billing via the API (and large companies having enough AI FOMO that they blindly pay the enormous invoices every month).

They are not losing money on subscription plans. Inference is very cheap - just a few dollars per million tokens. What they’re trying to do is bundle R&D costs with inference so they can fund the training of the next generation of models.

Banning third-party tools has nothing to do with rate limits. They’re trying to position themselves as the Apple of AI companies -a walled garden. They may soon discover that screwing developers is not a good strategy.

They are not 10× better than Codex; on the contrary, in my opinion Codex produces much better code. Even Kimi K2.5 is a very capable model I find on par with Sonnet at least, very close to Opus. Forcing people to use ONLY a broken Claude Code UX with a subscription only ensures they loose advantage they had.

> "just a few dollars per million tokens"

Google AI Pro is like $15/month for practically unlimited Pro requests, each of which take million tokens of context (and then also perform thinking, free Google search for grounding, inline image generation if needed). This includes Gemini CLI, Gemini Code Assist (VS Code), the main chatbot, and a bunch of other vibe-coding projects which have their own rate limits or no rate limits at all.

It's crazy to think this is sustainable. It'll be like Xbox Game Pass - start at £5/month to hook people in and before you know it it's £20/month and has nowhere near as many games.

OpenAI only released ChatGPT 4 years ago but…

Google has made custom AI chips for 11 years — since 2015 — and inference costs them 2-5x less than it does for every other competitor.

The landmark paper that invented the techniques behind ChatGPT, Claude and modern AI was also published by Google scientists 9 years ago.

That’s probably how they can afford it.

I agree that the TPUs are one of the things that are underestimated (based on my personal reading of HN).

Google already has a huge competitive advantage because they have more data than anyone else, bundle Gemini in each android to siphon even more data, and the android platform. The TPUs truly make me believe there actually could be a sort of monopoly on LLMs in the end, even though there are so many good models with open weights, so little (technical) reasons to create software that only integrates with Gemini, etc.

Google will have a lion‘s share of inferring I believe. OpenAI and Claude will have a very hard time fighting this.

I’m not familiar with the Claude Code subscription, but with Codex I’m able to use millions of tokens per day on the $200/mo plan. My rough estimate was that if I were API billing, it would cost about $50/day, or $1200/mo. So either the API has a 6x profit margin on inference, the subscription is a loss leader, or they just rely on most people not to go anywhere near the usage caps.

It's the latter. It's the average use that matters. Though I suspect API margins are also probably higher than people think.

I use GLM lite subscription for personal use. It is advertised as 3x claude code pro (the 20$ one).

5h allowance is somewhere between 50M-100M tokens from what I can tell.

On 200$ claude code plan you should be burning hundreds of millions of token per day to make anthropic hurt.

IMHO subscription plans are totally banking on many users underusing them. Also LLM providers dont like to say exact numbers (how much you get , etc)

Inference might be cheap, but I'm 100% sure Anthropic has been losing quite a lot of money with their subscription pricing with power users. I can literally see comparison between what my colleagues Claude cost when used with an API key vs when used with a personal subscription, and the delta is just massive

I wonder how many people have a subscription and don’t fully utilize it. That’s free money for them, too.

The trick is that the jump goes from 20 to 100 Dollar for the Pro to Max subscription. Pro is not enough for me, Max is too much. 60 would be ideal, but currently at 100 it's worth the cost.

But this is how every subscription works. Most people lose money on their gym subscription, but the convenience takes us.

What can bite them in this case though is alternate providers at the same price point that can bridge the gap. e.g. you currently get a lot more bang for your buck with the $20 OpenAI Codex subscription than you get for the $20 Claude Code subscription.

Of course they bundle R&D with inference pricing, how else could you the recoup that investment.

The interesting question is: In what scenario do you see any of the players as being able to stop spending ungodly amounts for R&D and hardware without losing out to the competitors?

In the scenario where that market collapses, ie when we stop making significant gains with new models. It might be a while, though, who knows.

Didn't OpenAI spend like 10 billion on inference in 2025? Which is around the same as their total revenue?

Why do people keep saying inference is cheap if they're losing so much money from it?

When you have 800–900 million active users, no matter how cheap it is, your costs will be in the billions.

They paid about $10B on inference and had about $10B in revenue in 2025. The users and numbers of zeroes on those numbers are not relevant. What is relevant is the ratio of those numbers. They apparently are not even profitable on inference, wich is the cheap part of the whole business.

And cost of inference tripled from $3B in 2024 to $10B in 2025, so cost of revenue linearly grows with number of users, i.e. it does not get cheaper.

https://www.wheresyoured.at/oai_docs/

What walled garden man? There’s like four major API providers for Anthropic.

For example, OpenAI’s agent (Codex) is open source, and you can use any harness you want with your OpenAI subscription. Anthropic keeps its tooling closed source and forbids using third-party tooling with a Claude subscription.

Except all those GPUs running inference need to be replaced every 2 years.

Why?

They wear down being run at 100% all the time. Support slowly drops off, the architecture and even the rack format become deprecated.

GPUs do not wear down from being ran at 100%, unless they're pushed past their voltage limits, or gravely overheating.

You can buy a GPU that's been used to mine bitcoin for 5 years with zero downtime, and as long as it's been properly taken care of (or better, undervolted), that GPU functions the exact same as a 5 year old GPU in your PC. Probably even better.

GPUs are rated to do 100%, all the time. That's the point. Otherwise it'd be 115%.

Yeah that's not how it works in practice in a datacenter with the latest GPUs, they are basically perishable goods.

You don't run your gaming PC 24/7.

No, you're fundamentally wrong. There's the regular wear & tear of GPUs that all have varying levels of quality, you'll have blown capacitors (just as you do with any piece of hardware), but running in a datacenter does not damage them more. If anything, they're better taken care of and will last longer. However, since instead of having one 5090 in a computer somewhere, you have a million of them. A 1% failure rate quickly makes a big number. My example included mining bitcoin because, just like datacenters, they were running in massive farms of thousands of devices. We have the proof and the numbers, running at full load with proper cooling and no over voltage does not damage hardware.

The only reason they're "perishable" is because of the GPU arms race, where renewing them every 5 years is likely to be worth the investment for the gains you make in power efficiency.

Do you think Google has a pile of millions of older TPUs they threw out because they all failed, when chips are basically impossible to recycle ? No, they keep using them, they're serving your nanobanana prompts.

GPU bitcoin mining rigs had a high failure rate too. It was quite common to run at 80% power to keep them going longer. That's before taking into account that the more recent generations of GPUs seems to be a lot more fragile in general.

Mining rigs also used more milk cartons than datacenter racks; [hot/cold] aisles? No, piles! Not to mention the often questionable power delivery...

AI data centers are also incentivised to reduce costs as far as they can. They could absolutely be running them in questionable setups

Yeah what's crazy is most of these companies are making accounting choices that obscure the true cost. By extending the stated useful life of their equipment, in some cases from 3 years to 6. Perfectly legal. And it has the effect of suppressing depreciation expenses and inflating reported earnings.

But don't they palpitate for thise sweet depreciation credits to decrease their tax on revenue?

Small sacrifice to not spook investors and the market.

> They are not losing money on subscription plans. Inference is very cheap - just a few dollars per million tokens. What they’re trying to do is bundle R&D costs with inference so they can fund the training of the next generation of models.

You've described every R&D company ever.

"Synthesizing drugs is cheap - just a few dollars per million pills. They're trying to bundle pharmaceutical research costs... etc."

There's plenty of legit criticisms of this business model and Anthropic, but pointing out that R&D companies sink money into research and then charge more than the marginal cost for the final product, isn't one of them.

I’m not saying charging above marginal cost to fund R&D is weird. That’s how every R&D company works.

My point was simpler: they’re almost certainly not losing money on subscriptions because of inference. Inference is relatively cheap. And of course the big cost is training and ongoing R&D.

The real issue is the market they’re in. They’re competing with companies like Kimi and DeepSeek that also spend heavily on R&D but release strong models openly. That means anyone can run inference and customers can use it without paying for bundled research costs.

Training frontier models takes months, costs billions, and the model is outdated in six months. I just don’t see how a closed, subscription-only model reliably covers that in the long run, especially if you’re tightening ecosystem access at the same time.

Yes, and my point is that thinking the cost of subscriptions is only inference, and not the research, is mistaken.

They can totally lose money on subscriptions despite the costs of inference, because research costs have to be counted too.

> Yes, and my point is that thinking the cost of subscriptions is only inference, and not the research, is mistaken.

Of course they are losing money when you factor in R&D. Everybody knows that. That is not what people mean when they say that they "lose money" on subscriptions.

"They're not losing money on subscriptions, it's just their revenue is smaller than their costs". Weird take.

It means the marginal cost to sell another subscription is lower than what they sell it for. I don't know if that's true, but it seems plausible.

The secret is there is no path on making that back.

the path is by charging just a bit less than the salary of the engineers they are replacing.

After hearing this 10 times a day for the last 5 years I'm starting to get a bit tired. Do you have a rough time for when this great replacement is coming? 1 year? 2? 5? If it's longer than that can we shut up about it for a few years please.

It’s happening already? Ask any new CS grads about how good the job market is.

A poor economy that is still dealing with a decade+ of ZIRP, COVID shock, tariffs, and political strife; I don't see how AI has much, if anything, to do with this when compared with other options.

If AI was truly this productive they wouldn't be struggling so hard to sell their wares.

My crude metaphor to explain to my family is gasoline has just been invented and we're all being lent Bentley's to get us addicted to driving everywhere. Eventually we won't be given free Bentley's, and someone is going to be holding the bag when the infinite money machine finally has a hiccup. The tech giants are hoping their gasoline is the one that we all crave when we're left depending on driving everywhere and the costs go soaring.

Why? Computers and anything computer related have historically been dropping in prices like crazy year after year (with only very occasional hiccups). What makes you think this will stop now?

Commodity hardware and software will continue to drop in price.

Enterprise products with sufficient market share and "stickiness", will not.

For historical precedent, see the commercial practices of Oracle, Microsoft, Vmware, Salesforce, at the height of their power.

> Commodity hardware and software will continue to drop in price.

The software is free (citation: Cuda, nvcc, llvm, olama/llama cpp, linux, etc)

The hardware is *not* getting cheaper (unless we're talking a 5+ year time) as most manufacturers are signaling the current shortages will continue ~24 months.

> The hardware is not getting cheaper (unless we're talking a 5+ year time)

Yes, that's the time I'm talking about.

You also had a blip with increasing hard disk prices when Thailand flooded a few years ago.

GB300 NVL72 is 50% more expensive than GB200 I've heard.

It has stopped. Demand is now rising faster than supply in memory, storage and GPUs.

We see vendors reducing memory in new smart phones in 2026 vs 2025 for example.

At least for the moment falling consumer tech hardware prices are over.

Memory and storage has always been very cyclical. This is nothing new

On consumer side looking at a few past generations I question that. I would guess that we are nearing some sort of plateau there or already on it. There was inflation, but still not even considering RAM prices from last jump gains relative to cost were not that massive.

Recent price trends for DRAM, SSDs, hard drives?

Short term squeeze, because building capacity takes time and real funding. The component manufacturers have been here before. Booms rarely last long enough to justify a build-out. If AI demand turns out to be sustained, the market will eventually adapt by building supply, and prices will drop. If AI demand turns out to be transient, demand will drop, and prices will drop.

Cars have also been dropping in price.

And knives apparently.

I recently encountered this randomly -- knives are apparently one of the few products that nearly every household has needed since antiquity, and they have changed fairly little since the bronze age, so they are used by economists as a benchmark that can span centuries.

Source: it was an aside in a random economics conversation with charGPT (grain of salt?).

There is no practical upshot here, but I thought it was cool.

Yeah I’d definitely take that knife thing with a grain of salt. I have most of a history degree, took a lot of Econ classes (before later going back for CS), and it’s a topic I’m very interested in and I’ve never heard that (and some digging didn’t find anything).

It’s also false that the technology has changed very little.

The jumps from bronze to iron to steel to modern steel and sometimes to stainless steel all result in vastly different products. Not to mention the advances in composite materials for handles.

Then you need to look at substitute goods and the what people actually used knives for.

A huge amount of the demand for knives evaporated thanks to societal changes and substitute goods like forks. A few hundred years ago the average person had a knife that was their primary eating utensil, a survival tool, and a self defense weapon. Knives like that exist today but they’re not something every household has or needs.

This is a good example of why learning from ChatGPT is dangerous. This is a story that sounds very plausible at first glance, but doesn’t make sense once you dig in.

Evidence for this claim?

I had a 1990 Ford Taurus as my first car. I had got it used and I remember it being completely impossible to afford a new car at the time.

It was sticker price of $33,000 adjusted for inflation:

https://en.wikipedia.org/wiki/Ford_Taurus_%28second_generati...

I don't think it would even feel safe to drive at all compared to what we have got use to with modern cars. It broke down 3 times while I had it and stranded me on the road. No cell phone of course to call anyone.

These were the mythic "good ol days".

A few generations ago almost nobody could afford a car, now many low income families afford two.

Please show me where any AI company is currently turning a profit with their current offering and price structure, then let's have that conversation.

In the GP's analogy, the Bentley can be rented for $3/day, but if you want to purchase it outright, it will cost you $3,000,000.

Despite the high price, the Bentley factory is running 24/7 and still behind schedule due to orders placed by the rental-car company, who has nearly-infinite money.

I like this analogy.

I also think we're, as ICs, being given Bentleys meanwhile they're trying to invent Waymos to put us all out of work.

Humans are the cost center in their world model.

[dead]

how do I understand what is the sustainable pricing?

Depends on how you do the accounting. Are you counting inference costs or are you amortizing next gen model dev costs. "Inference is profitable" is oft repeated and rarely challenged. Most subscription users are low intensity users after all.

I agree; unfortunately when I brought up that they're losing before I get jumped on demanding me to "prove it" and I guess pointing at their balance sheets isn't good enough.

The question I have: how much are they _also_ losing on per-token billing?

From what I understand, they make money per-token billing. Not enough for how much it costs to train, not accounting for marketing, subscription services, and research for new models, but if they are used, they lose less money.

Finance 101 tldr explanation: The contribution margin (= price per token -variable cost per token ) this is positive

Profit (= contribution margin x cuantity- fix cost)

Do they make enough to replace their GPUs in two years?

if 100% of the money they spend is in inference priced by tokens (they don't say about subscriptions so i asume they lost money), yes they make money, but their expenses are way higher than inference alone. so they can make the gpu cost if they sell tokens but in reality this isnt the case, becouse they have to constaly train new models, subscription marketing, R&D, And overhead. antropic in general lost way less money than their competitors i will take this number in particular the projected break even but googling say Gross margin in this case is how much money they do whit the GPU " Gross Margins: Projected to swing from negative 94% last year to as much as 50% this year, and 77% by 2028. Projected Break-even: The company expects to be cash flow positive by 2027 or 2028. "

i will not be as bullish to say they will no colapse (0 idear how much real debt and commitments they have, if after the bubble pop spending fall shraply, or a new deepseek moment) but this sound like good trajectory (all things considered) i heavily doubt the 380 billions in valuation

"this is how much is spendeed in developers between $659 billion and $737 billion. The United States is the largest driver of this spending, accounting for more than half of the global total ($368.5 billion in 2024)" so is like saying that a 2% of all salaries of developers in the world will be absorbed as profit whit the current 33.3 ratio, quite high giving the amount of risk of the company.

Why do you think they're losing money on subscriptions?

Does a GPU doing inference server enough customers for long enough to bring in enough revenue to pay for a new replacement GPU in two years (and the power/running cost of the GPU + infrastructure). That's the question you need to be asking.

If the answer is not yes, then they are making money on inference. If the answer is no, the market is going to have a bad time.

Because they're not saying they are making a profit

That doesn’t mean that the subscription itself is losing money. The margin on the subscription could be fine, but by using that margin to R&D the next model, the org may still be intentionally unprofitable. It’s their investment/growth strategy, not an indictment of their pricing strategy.

They have investors that paid for training of these models too. It could be argued that R&D for the next generation is a separate issue, but they need to provide a return on the R&D in this generation to stay in business.

The return of R&D can just be an inflated valuation, there's no immediate need to make actual money.

But why does it matter which program you use to consume the tokens?

The sounds like a confession that claude code is somewhat wasteful at token use.

No, it's a confession they have no moat other than trying to hold onto the best model for a given use case.

I find that competitive edge unlikely to last meaningfully in the long term, but this is still a contrarian view.

More recently, people have started to wise up to the view that the value is in the application layer

https://www.iconiqcapital.com/growth/reports/2026-state-of-a...

But if more users use your service you get an advantage. Letting the users chose their tool for that would be a good thing.

Honestly I think I am already sold on AI, who is the first company that is going to show us all how much it really costs and start enshitification? First to market wins right?

Not according to this guy who works on Claude Code: https://x.com/trq212/status/2024212378402095389?s=20

What a PR nightmare, on top of an already bad week. I’ve seen 20+ people on X complaining about this and the related confusion.

No, it is prohibited. They're just updating the docs to be more clear about their position, which haven't changed. Their docs was unclear about it.

Yes, it was always prohibited, hence the OpenCode situation one or two months ago.

They really need to correct that. I understand jack shit. Is openclaw banned under these terms? Or just abuse where I build a business on top of that? And why does it matter anyway? I have my token restrictions ... So let me do what I want.

woof, does Anthropic not have a comms team and a clear comms policy for employees that aren’t on that comms team?

Probably not, they’re like four years old and they’re 2500 people at the company. My guess is that there are but a handful of PMs.

Incorrect, the third-party usage was already blocked (banned) but it wasn't officially communicated or documented. This post is simply identifying that official communication rather than the inference of actual functionality.

I pay a Max subscription since a long time, I like their model but I hate their tools:

- Claude Desktop looks like a demo app. It's slow to use and so far behind the Codex app that it's embarassing.

- Claude Code is buggy has hell and I think I've never used a CLI tool that consume so much memory and CPU. Let's not talk about the feature parity with other agents.

- Claude Agent SDK is poorly documented, half finished, and is just thin wrapper around a CLI tool…

Oh and none of this is open source, so I can do nothing about it.

My only option to stay with their model is to build my own tool. And now I discover that using my subscription with the Agent SDK is against the term of use?

I'm not going to pay 500 USD of API credits every months, no way. I have to move to a different provider.

I agree that Claude Code is buggy as hell, but:

> Let's not talk about the feature parity with other agents.

What do you mean feature parity with other agents? It seems to me that other CLI agents are quite far from Claude Code in this regard.

Which other CLI agents are that? Because I've found OpenCode to be A LOT better than Claude-Code.

Whats better with opencode? Never tried it. I like that claude code has double escape, shift + tab, team of agents

> Claude Code is buggy has hell and I think I've never used a CLI tool that consume so much memory and CPU

FWIW this aligns completely with the LLM ethos. Inefficiency is a virtue.

I had a Claude code instance using 55 GB of RAM yesterday.

I got so tired of cursor that I started writing down every bug I encountered. The list is currently at 30 entries, some of them major bugs such as pressing "apply" on changes not actually applying changes or models getting stuck in infinite loops and burning 50 million tokens.

I tried to have Cursor change a list of US States and Provinces from a list to a dictionary and it did, but it also randomly deleted 3 states.

I regret ever promoting that Claude Code crap. I remember when it was nothing but glowing reviews everywhere. Honestly AI companies should stick to what they are good at: direct API interface to powerful models.

We are heading toward a $1000/month model just to use LLMs in the cloud.

Your core customers are clearly having a blast building their own custom interfaces, so obviously the thing to do is update TOS and put a stop to it! Good job lol.

I know, I know, customer experience, ecosystem, gardens, moats, CC isn't fat, just big boned, I get it. Still a dick move. This policy is souring the relationship, and basically saying that Claude isn't a keeper.

I'll keep my eye-watering sub for now because it's still working out, but this ensures I won't feel bad about leaving when the time comes.

Update: yes yes, API, I know. No, I don't want that. I just want the expensive predictable bill, not metered corporate pricing just to hack on my client.

They'll all do this eventually.

We're in the part of the market cycle where everyone fights for marketshare by selling dollar bills for 50 cents.

When a winner emerges they'll pull the rug out from under you and try to wall off their garden.

Anthropic just forgot that we're still in the "functioning market competition" phase of AI and not yet in the "unstoppable monopoly" phase.

"Naveen Rao, the Gen AI VP of Databricks, phrased it quite well:

all closed AI model providers will stop selling APIs in the next 2-3 years. Only open models will be available via APIs (…) Closed model providers are trying to build non-commodity capabilities and they need great UIs to deliver those. It's not just a model anymore, but an app with a UI for a purpose."

~ https://vintagedata.org/blog/posts/model-is-the-product A. Doria

> new Amp Free (10$) access is also closed up since of last night

Unstoppable monopoly will be extremely hard to pull off given the number of quality open (weights) alternatives.

I only use LLMs through OpenRouter and switch somewhat randomly between frontier models; they each have some amount of personality but I wouldn't mind much if half of them disappeared overnight, as long as the other half remained available.

I'm old, so I remember saying the same thing about Google and search.

I hope you're right!

I think the big difference is that Google is free: everyone is using Google because it doesn’t cost anything and for a long time was the best search engine out there. I am sure that if Google would suddenly charge a few dollars per month for access, Bing market share would explode overnight, because it would become “good enough but cheaper”.

With the AI models, using a model that is “good enough but cheaper” is already an option.

There's no reason that a sizeable portion of LLM usage can't and won't end up free/ad-sponsored. Cutting edge stuff for professional use will probably be monetized via subscription or API credits for a long time to come. But running an older and less resource intensive model works just fine for tasks like summarization. These models will just become another feature in a "free" product that people pay for by watching or clicking ads.

I imagine the split will look a lot like b2b vs b2c in other technologies, b2b customers tend to be willing to pay for tech when it offers a competitive advantage, reduces their operating costs etc. b2c customers mostly just guzzle free slop.

It's actually pretty bonkers when you think about how basically every cutting edge professional you deal with is getting ads for all of their top search results for all of their work.

(Not quite "every", but outside of tech, most professional workplaces don't support ad blocking or Kagi.)

I too am old. Google search is free, hard to replicate, and while there used to be lots of search engines, Google was (and arguably still is) miles ahead of all the others in terms of quality and performance.

A model is hard to train but it doesn't need to be hyper up to date / have a new version come out every day. Inference is cheap (it seems?) and quality is comparable. So it's unclear how expensive offerings could win over free alternatives.

I could be wrong of course. I don't have a crystal ball. I just don't think this is the same as Google.

Of course I could be entirely mistaken and there could emerge a single winner

I would say Google's monopoly mainly comes from its name recognition, definitely not because its still ahead in core search as I have been using DuckDuckGo for 2 years once I noticed search results are the same or better than Google.

In the first years, I remember no other search engine was close to Google quality. We all ditched AltaVista because Google was incredibly better. It would have been awful to switch back to any other options. We can already switch between the 3 big proprietary models without feeling too much differences, so it’s quite a different landscape.

Yes, my point exactly.

This is saying we have hundreds of open source OSes and Windows will never be a monopoly.

Software always gets monopoly simply by usage. Every time a model gets used by esoteric use cases, it gets more training data (that a decentralized open weight model doesn't get) and it starts developing its moat.

I think windows has historical monopoly.

They bundled it with PC hw and the vast majority of apps only ever got published for windows, and this over decades (one would argue it’s still true).

The starting point for LLMs is very different. Who would publish today a software that only integrates with chatGPT? Only a small minority.

Thus I agree, I struggle to see how a monopoly can exist here. A GPU monopoly or duopoly though, perhaps.

> Software always gets monopoly simply by usage

Most software isn't made by monopolies. More directly, enterprise-software stocks are getting hammered because AI offers them competition.

It's more like saying AWS has a monopoly on virtual machine hosting.

(For those unaware, AWS doesn't have a VM monopoly, and the market dynamics seem similar)

It’ll be a bunch of tiny moats in that scenario. LLMs are way too generic, adaptable, flexible in how you use it to make a big most out of it.

>This is saying we have hundreds of open source OSes

we don't, we have about 3 operating systems that have the decades of hardware and software compatibility that makes them widely usable. They're the most complex and complicated things we've built. LLMs are a few thousand lines of python hooked up to a power plant and graphics cards. This is the least defensible piece of software there ever has been.

OpenRouter falls in the acceptable use category. They targeting users that are misusing their Claude OAuth token on non-Anthropic products.

They will [try to] ban open weights for ethics / security reasons: to stop spammers, to protect children, to stop fascism, to defend minorities. Take your pick; it won't matter why, it will only matter which media case can they thrust in the spotlight first.

Yes of course they will; the CEO of Anthropic makes that argument, very openly, all the time. But it will be hard to do, I think.

Hopefully, but they'll lobby hard, supported by all the money they raised.

> They'll all do this eventually

And if the frontier continues favouring centralised solutions, they'll get it. If, on the other hand, scaling asymptotes, the competition will be running locally. Just looking at how much Claude complains about me not paying for SSO-tier subscriptions to data tools when they work perfectly fine in a browser is starting to make running a slower, less-capable model locally competitive with it in some research contexts.

Imagine having a finite pool of GPUs worth more than their weight in gold, and an infinite pool of users obsessed with running as many queries against those GPUs in parallel as possible, mostly to review and generate copious amounts of spam content primarily for the purposes of feeling modern, and all in return for which they offer you $20 per month. If you let them, you must incur as much credit liability as OpenAI. If you don't, you get destroyed online.

It almost makes me feel sorry for Dario despite fundamentally disliking him as a person.

Hello old friend, I've been expecting you.

First of all, custom harness parallel agent people are so far from the norm, and certainly not on the $20 plan, which doesn't even make sense because you'd hit token limit in about 90 seconds.

Second, token limits. Does Anthropic secretly have over-subscription issues? Don't know, don't care. If I'm paying a blistering monthly fee, I should be able to use up to the limit.

Now I know you've got a clear view of the typical user, but FWIW, I'm just an aging hacker using CC to build some personal projects (feeling modern ofc) but still driving, no yolo or gas town style. I've reached the point where I have a nice workflow, and CC is pretty decent, but it feels like it's putting on weight and adding things I don't want or need.

I think LLMs are an exciting new interface to computers, but I don't want to be tied to someone else's idea of a client, especially not one that's changing so rapidly. I'd like to roll my own client to interface with the model, or maybe try out some other alternatives, but that's against the TOS, because: reasons.

And no, I'm not interested in paying metered corporate rates for API access. I pay for a Max account, it's expensive, but predictable.

The issue is Anthropic is trying for force users into using their tool, but that's not going to work for something so generic as interfacing with an LLM. Some folks want emacs while others want vim, and there will never be a consensus on the best editor (it's nvim btw), because developers are opinionated and have strong preferences for how they interface with computers. I switched to CC maybe a year ago and haven't looked back, but this is a major disappointment. I don't give a shit about Anthropic's credit liability, I just want the freedom to hack on my own client.

You're not "rolling your own client." You're using a subscription that prices in a specific usage pattern, the one mediated by their client, and trying to route around it to extract more value than you're paying for. That's not hacking, it's arbitrage, and pretending it's about editor philosophy is cope.

Anthropic sells two products: a consumer subscription with a UI, and an API with metered pricing. You want the API product at the subscription price. That's not a principled stance about interface freedom, it's just wanting something for less than it costs.

The nvim analogy doesn't land either. Nobody's stopping you from writing your own client. You just have to pay API rates for it, because that's the product that matches what you're describing. The subscription subsidises the cost per token by constraining how you use it. Remove the constraint, the economics break. This isn't complicated.

"I don't give a shit about Anthropic's credit liability," right, but they do, because it's their business. You're not entitled to a flat-rate all-you-can-eat API just because you find metered pricing aesthetically displeasing.

As my sibling mentioned, it's not all-you-can-eat, it's metered and capped. Is it the client mediating usage patterns? I don't know, but why not do it server-side and let people use it however they want? Control of course, but that's not in the interest of users.

I'm not trying to arbitrage or route around anything, I just want predictable billing to access the model. Maybe the API would be cheaper for me, I don't know. I'm just a normal user, not scheduling an agent army to blast at maximum 24/7.

You don't need to explain what Anthropic is selling, I get it, but you're off-base claiming that I'm pretending about editor philosophy as cope. I think Anthropic is where they are today because they have a good model for coding, made popular by software folks who value things like editor choice and customization. Anthropic is free to do as they wish of course, as am I, but I'm displeased with their decision here, and voicing my opinion about it.

If usage is truly constrained by the client not the server, I guess I can't argue that, but it still feels bad as an end user. As a consumer, I just want a fair deal and freedom to use what I purchase however I see fit. But that seems harder to find these days, and most businesses seem intent on maximum extraction by any means possible. I might be wrong, but this feels like business move to build a consumer moat by controlling the interface, because consumers don't want the API. It's not in my best interests, which alienates me as a customer.

It seems to me the consumer subscription still has metered token limits (it's not all you can eat), so why should it matter how we use those tokens?

Why do you fundamentally dislike him as a person?

The only thing I've seen from him that I don't like is the "SWEs will be replaced" line (which is probably true and it's more that I don't like the factuality of it).

It’s kinda obvious he’s a well spoken shark. Personally not an issue for me, you have to be at the top of a unicorn, but it isn’t something people in general like.

Interesting, are there any sources for the shark claims ? I recently saw an interview with Hassabis and him and thought: <well at least those are two actual scientist leading AI labs/devisions>, so that gave me some hope that what they discussed regarding security and eventual equal distribution of "AI" benefits had some genuine intention.

Don’t be mad at it, be happy you were able to throw some of that sweet free vc money at your hobbies instead of paying the market rate.

Oh I'm not mad, it's more of a sad clown type of thing. I'm still stoked to use it for now. We can always go back to the old ways if things don't work out.

They offer an API for people who want to build their own clients. They didn't stop people from being able to use Claude.

at a significantly higher price... which of course is why they're doing this.

That's what the API is for.

So basically you are saying Anthropic models are indispensable but you are too cheap to pay for it.

Nowhere did I say they're indispensable, and I explicitly said I'm still paying for it. If all AI companies disappear tomorrow that's fine. I'm just calling out what I think is tone-deaf move, by a company I pay a large monthly bill to.

Sure they are having a blast, they are paying 20$ instead of getting charged hundreds for forr tokens.

It's simple, follow the ToS

This is funny. This change actually pushes me into using a competitor more (https://www.kimi.com). I was trying out this provider with oh-my-pi (https://github.com/can1357/oh-my-pi) and was lamenting that it didn't have web search implemented using kimi.

Well a kind contributor just added that feature specifically because of this ban(https://github.com/can1357/oh-my-pi/pull/110).

I'm happy as a clam now. Yay for competition!

Is it me, or will this just speed up the timeline where a 'good enough' open model (Qwen? Deepseek? - I'm sure the Chinese will see a value in undermining OpenAI/Anthropic/Google) combined with good enough/cheap hardware (10x inference improvement in a M7 Macbook Air?) makes running something like opencode code locally a no brainer?

The good enough alternative models are here or will be soon, depending on your definition of good enough. MiniMax-M2.5 looks really competitive and its a tenth of the cost of Sonnet-4.6 (they also have subscriptions).

Running locally is going to require a lot of memory, compute, and energy for the foreseeable future which makes it really hard to compete with ~$20/mo subscriptions.

Personally I am already there- I go to Qwen and Deepseek locally via ollama for my dumb questions and small tasks, and only go to Claude if they fail. I do this partially because I am just so tired of everything I do over a network being logged, tracked, mined and monetized, and also partially because I would like my end state to be using all local tools, at least for personal stuff.

People running models locally has always been the scare for the sama's of the world. "Wait, I don't need you to generate these responses for me? I can get the same results myself?"

He can't buy all the RAM

Going to keep using the agents sdk with my pro subscription until I get banned. It's not openclaw it's my own project. It started by just proxying requests to claude code though the command line, the sdk just made it easier. Not sure what difference it makes to them if I have a cron job to send Claude code requests or an agent sdk request. Maybe if it's just me and my toy they don't care. We'll see how the clarify tomorrow.

There is a new breed of agent-agnostic tools that call the Claude Code CLI as if it's an API (I'm currently trying out vibe-kanban).

This could be used to adhere to Claude's TOS while still allowing the user to switch AI companies at a moment's notice.

Right now there's limited customizability in this approach, but I think it's not far-fetched to see FAR more integrated solutions in the future if the lock-in trend continues. For example: one MCP that you can configure into a coding agent like Claude Code that overrides its entire behavior (tools, skills, etc.) to a different unified open-source system. Think something similar to the existing IntelliJ IDEA's MCP that gives a separate file edit tool, etc. than the one the agent comes with.

Illustration of what i'm talking about:

- You install Claude Code with no configuration

- Then you install the meta-agent framework

- With one command the meta-agent MCP is installed in Claude Code, built-in tools are disabled via permissions override

- You access the meta-agent through a different UI (similar to vibe-kanban's web UI)

- Everything you do gets routed directly to Claude Code, using your Claude subscription legally. (Input-level features like commands get resolved by meta-agent UI before being sent to claude code)

- Claude Code must use the tools and skills directly from meta-agent MCP as instructed in the prompt, and because its own tools are permission denied (result: very good UI integration with the meta-agent UI)

- This would also work with any other CLI coding agent (Codex, Gemini CLI, Copilot CLI etc.) should they start getting ideas of locking users in

- If Claude Code rug-pulls subscription quotas, just switch to a competitor instantly

All it requires is a CLI coding agent with MCP support, and the TOS allowing automatic use of its UI (disallowing that would be massive hypocrisy as the AI companies themselves make computer use agents that allow automatic use of other apps' UI)

I think I've made two good decisions in my life. The first was switching entirely to Linux around '05 even though it was a giant pain in the ass that was constantly behind the competition in terms of stability and hardware support. It took awhile but wow no regrets.

The second appears to be hitching my wagon to Mistral even though it's apparently nowhere as powerful or featureful as the big guys. But do you know how many times they've screwed me over? Not once.

Maybe it's my use cases that make this possible. I definitely modified my behavior to accommodate Linux.

They're too small to screw you over. But you've got more time until they do at least.

[dead]

For what it's worth, I built an alternative specifically because of the ToS risk. GhostClaw uses proper API keys stored in AES-256-GCM + Argon2id encrypted vault -no OAuth session tokens, no subscription credentials, no middleman. Skills are signed with Ed25519 before execution. Code runs in a Landlock + seccomp kernel sandbox. If your key gets compromised you rotate it; if a session token gets compromised in someone else's app you might not even know.

t's open source, one Rust binary, ~6MB. https://github.com/Patrickschell609/ghostclaw

AI is the new high-end gym membership. They want you to pay the big fee and then not use what you paid for. We'll see more and more roadblocks to usage as time goes on.

This was the analogy I was looking for! It feels like a very creepy way to make money, almost scammy and the gym membership/overselling hits the nail.

This feels more like the gym owner clarifying it doesn't want you using their 24-hour gym as a hotel just because you find their benches comfortable to lie down on, rather than a "roadblock to usage"

Not really, these subscriptions have a clear and enforced 5h and weekly limit.

Sorry but if you're not paying the big fee there's no way you're going to have a job by the late-2020s.

The fundamental tension here is that AI companies are selling compute at a loss to capture market share, while users are trying to maximize value from their subscriptions.

From a backend perspective, the subscription model creates perverse incentives. Heavy users (like developers running agentic workflows) consume far more compute than casual users, but pay the same price. Third-party tools amplify this asymmetry.

Anthropic's move is economically rational but strategically risky. Models are increasingly fungible - Gemini 3.1 and Claude 4.5 produce similar results for most tasks. The lock-in isn't the model; it's the tooling ecosystem.

By forcing users onto Claude Code exclusively, they're betting their tooling moat is stronger than competitor models. Given how quickly open-source harnesses like pi have caught up, that's a bold bet.

This is how you gift wrap the agentic era to the open source chinese LLMs. devs don't need the best model, they need one without lawyers attached.

The pressure is to boost revenue by forcing more people to use the API to generate huge numbers of tokens they can charge more for. LLMs are becoming common commodities as open weight models keep catching up. There are similarities with pirating in the 90s when users realize they can ctrl+c ctrl+v to copy a file/model and you don't need to buy a cd/use their paid API.

And that is how it should be - the knowledge that the LLM trained on should be free, and cannot (and should never be) gatekept behind money.

It's merely the hardware that should be charged for - which ought to drop in price if/when the demand for it rises. However, this is a bottleneck at the moment, and hard to see how it gets resolved amidst the current US environment on sanctioning anyone who would try.

Is there no value in how the training was done such that it's accessible via inference in a particularly useful way?

That value is there, but google has decided to give it away as public knowledge (ala, their transformer paper).

And i would also argue that the researchers doing this are built on shoulders of other public knowledge - things funded by public institutions with taxpayer money.

No, a lot of the data they were trained on was pirated.

This feels perfectly justifiable to me. The subscription plans are super cheap and if they insist you use their tool I understand. Ya'll seem a bit entitled if I'm being honest.

I just cancelled my Pro subscription. Turns out that Ollama Cloud with GLM-5 and qwen-coder-next are very close in quality to Opus, I never hit their rate limits even with two sessions running the whole day and there zero advantage for me to use Claude Code compared to OpenCode.

Anthropic is dead. Long live open platforms and open-weight models. Why would I need Claude if I can get Minimax, Kimi, and Glm for the fraction of the price?

To get comparable results you need to run those models on at least prosumer hardware and it seems that two beef-up Mac Studios are the minimum. Which means that instead of buying this hardware you can purchase Claude, Codex and many other subscriptions for next 20 years.

Or you purchase a year's worth of almost unlimited MiniMax coding plan for a price you'd pay for 15 days of limited Claude usage.

And as a bonus, you can choose your harness. You don't have to suffer CC.

And if something better appears tomorrow, you switch your model, while still using your harness of choice.

And do you think Claude won't have released a few models by then, which will be way ahead of any open model? you are paying for SOTA.

Being 3-6 months behind SOTA may be less of a hindrance than not being able to use the tools you want, though.

there’s a million small scale AI apps that just aren’t worth building because there’s no way to do the billing that makes sense. If anthropic wanted to own that market, they could introduce a bring-your-own-Claude metaphor, where you login with Claude and token costs get billed to your personal account (after some reasonable monthly freebies from your subscription).

But the big guys don’t seem interested in this, maybe some lesser known model will carve out this space

This is going to happen. Unfortunately.

I shudder to think what the industry will look like if software development and delivery becomes like Youtubing, where the whole stack and monetization is funneled through a single company (or a couple) get to decide who gets how much money.

I am a bit worried that this is the situation I am in with my (unpublished) commercial app right now: one of the major pain points I have is that while I have no doubt the app provides value in itself, I am worried about how many potential users will actually accept paying inference per token...

As an independent dev I also unfortunately don't have investors backing me to subsidize inference for my subscription plan.

I recommend kimi. It's possible for people to haggle with it to get cheap for the first month and as such try out your project and best part of the matter is that kimi intentionally supports api usage in any of their subscribed plan and they also recently changed their billing to be more token usage based like others instead of their previous tool calling limits

It's seriously one of the best models. very comparable to sonnet/opus although kimi isn't the best in coding. I think its a really great solid model overall and might just be worth it in your use case?

Is the use case extremely coding intensive related (where even some minor improvement can matter for 10-100x cost) or just in general. Because if not, then I can recommend Kimi.

Kimi terms prohibit enterprise use for their sub plans.

Please correct me if you feel I'm wrong after reading it.

>> there’s a million small scale AI apps that just aren’t worth building because there’s no way to do the billing that makes sense

Maybe they are not worth building at all then. Like MoviePass wasn’t.

I got banned for violating terms of use apparently, but I'm mystified as to what I rule I broke, and appealing just vanishes into the ether.

Two accounts of mine were banned for some reason and my sub was refunded. Literally from just inane conversations. Conversations also disappear and break randomly, but this happens on ChatGPT too sometimes

In enterprise software, this is an embedded/OEM use case.

And historically, embedded/OEM use cases always have different pricing models for a variety of reasons why.

How is this any different than this long established practice?

It's not, but do you really think the people having Claude build wrappers around Claude were ever aware of how services like this are typically offered.

From the legal docs:

> Authentication and credential use

> Claude Code authenticates with Anthropic’s servers using OAuth tokens or API keys. These authentication methods serve different purposes:

> OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.

> Developers building products or services that interact with Claude’s capabilities, including those using the Agent SDK, should use API key authentication through Claude Console or a supported cloud provider. Anthropic does not permit third-party developers to offer Claude.ai login or to route requests through Free, Pro, or Max plan credentials on behalf of their users.

> Anthropic reserves the right to take measures to enforce these restrictions and may do so without prior notice.

why wouldn't they just make it so the SDK can't use claude subs? like what are they doing here?

When your company happens upon a cash cow, you can either become a milk company or a meat company.

OK I hope someone from anthropic reads this. Your API billing makes it really hard to work with it in India. We've had to switch to openrouter because anthropic keeps rejecting all the cards we have tried. And these are major Indian banks. This has been going on for MONTHS

It’s the same here in Hong Kong. I can’t use any of my cards (personal or corporate) for OpenAI or Anthropic.

Have to do everything through Azure, which is a mess to even understand.

Thariq has clarified that there are no changes to how SDK and max suscriptions work:

https://x.com/i/status/2024212378402095389

---

On a different note, it's surprising that a company that size has to clarify something as important as ToS via X

> On a different note, it's surprising that a company that size has to clarify something as important as ToS via X

Countries clarify nation policy on X. Seriously it feels like half of the EU parliament live on twitter.

FYI a Twitter post that contradicts the ToS is NOT a clarification.

What's wrong with using X?

In the case you are asking in good faith, a) X requires logging in to view most of its content, which means that much of your audience will not see the news because b) much of your audience is not on X, either due to not having social media or have stopped using X due to its degradation to put it generally.

I'm not signed in but I can view the above linked tweet just fine.

Plus it's not a real clarification in anyway. It's just PR. Even if it's posted on Mastodon or Github or anywhere, I highly doubt you can use it to defend yourself if you get banned from violating their ToS.

You can’t view answers and the tweet threat. You need to know every single tweet. You can’t open the politician‘s feed so you have to know that there is a tweet and which it is to get information.

c) it is controlled by a direct competitor and can bury / promote your customer communication at will.

Elon has enough sense not to cross that particular bridge.

He was quick to ban links to Mastodon when it was on the rise, I'm not sure why he'd treat SpaceX/xAI competitors any differently.

He's few borders behind that bridge now. They've been injecting faults left and right, from hiding tweets and accounts as "unavailable" to sorting replies by spamminess and everything.

Not bad per se but how much legal weight does it actually carry?

I presume zero.. but nonetheless seems like people will take it as valid anyway.

That can be dangerous I think.

And because of this i'll obviously opt to not subscribe to a Claude plan, when i can just use something like Copilot and use the models that way via OpenCode.

how comparable are the usage limits?

I feel like they want to be like Apple, and open-code + open-source models are Linux. The thing is, Apple is (for some) way better in user experience and quality. I think they can pull it off only if they keep their distance from the others. But if Google/Chinese models become as good as Claude, then there won’t be a reason — at least for me — to pay 10x for the product

At this point, where Kimi K2.5 on Bedrock with a simple open source harness like pi is almost as good the big labs will soon have to compete for users,... openai seems to know that already? While anthropic bans bans bans

Do you know by any chance if Bedrock custom model import also works with on-demand use, without any provisioned capacity? I'm still puzzled why they don't offer all qwen3 models on Bedrock by default.

I see a lot of Qwen3 in us west 2 And i have no experience with custom model on bedrock

I would expect, it still is only enforced in a semi-strict way.

I think what they want to achieve here is less "kill openclaw" or similar and more "keep our losses under control in general". And now they have a clear criteria to refer when they take action and a good bisection on whom to act on.

In case your usage is high they would block / take action. Because if you have your max subscription and not really losing them money, why should they push you (the monopoly incentive sounds wrong with the current market).

Openclaw is unaffected by this as the Claude Code CLI is called directly

Many people use the Max subscription OAuth token in OpenClaw. The main chat, heartbeat, etc., functionality does not call the Claude Code CLI. It uses the API authenticated via subscription OAuth tokens, which is precisely what Anthropic has banned.

There are many other options too: direct API, other model providers, etc. But Opus is particularly good for "agent with a personality" applications, so it's what thousands of OpenClaw users go with, mostly via the OAuth token, because it's much cheaper than the API.

This is a signal that everyone making AI apps should build on Gemini/OpenAI, and since there is a dance of code and model to get good results, inevitably Anthropic are now writing themselves out of being the backend for everyone elses AI apps going forward

The telemetry from claude code must be immensely valuable for training. Using it is training your replacement!

That page is... confusing.

> Advertised usage limits for Pro and Max plans assume ordinary, individual usage of Claude Code and the Agent SDK.

This is literally the last sentence of the paragraph before the "Authentication and credential use"

OpenClaw, NanoClaw, et al all use AgentSDK which will from now on be forbidden.

They are literally alienating a large percentage of OpenClaw, NanoClaw, PicoClaw, customers because those customers will surely not be willing to pay API pricing, which is at least 6-10x Max Plan pricing (for my usage).

This isn’t too surprising to me since they probably have a direct competitor to openclaw et al in the works right now, but until then I am cancelling my subscription and porting my nanoclaw fork with mem0 integration to work with OpenAI instead.

Thats not a “That’ll teach ‘em” statement, it is just my own cost optimization. I am quite fond of Anthropic’s coding models and might still subscribe again at the $20 level, but they just priced me out for personal assistant, research, and 90% of my token use case.

What does Anthropic have to gain from users who use a very high amount of tokens for OpenClaw, NanoClaw etc and pay them only $20?

Their moat is evaporating before our eyes. Anthropic is Microsoft's side piece, but Microsoft is married with kids to OpenAI.

And OpenAI just told Microsoft why they shouldn't be seeing Anthropic anymore; Gpt-5.3-codex.

RIP Anthropic.

Is this a direct shot at things like OpenClaw, or am I reading it wrong?

They even block Claude Code of you've modified it via tweakcc. When they blocked OpenCode, I ported a feature I wanted to Claude Code so I could continue using that feature. After a couple days, they started blocking it with the same message that OpenCode gets. I'm going to go down to the $20 plan and shift most of my work to OpenAI/ChatGPT because of this. The harness features matter more to me than model differences in the current generation.

Opencode as well. Folks have been getting banned for abusing the OAuth login method to get around paying for API tokens or whatever. Anthropic seems to prefer people pay them.

its not that innocent.

a 200 dollar a month customer isn't trying to get around paying for tokens, theyre trying to use the tooling they prefer. opencode is better in a lot of ways.

tokens get counted and put against usage limits anyway, unless theyre trying to eat analytics that are CC exclusive they should allow paying customers to consume to the usage limits in however way they want to use the models.

Anthropic is offering a steep discount in their plans. I highly doubt they want you using it in a harness where you can trivially switch away when someone else releases a better model

Funny, because you CAN switch Claude Code to other providers and models easily.

Anthropic is just a deeply "mis-dev-anthropic" company.

> they should allow paying customers to consume to the usage limits in however way they want to use the models.

I think I agree, but it's their business to run however they like. They have competition if we don't like it.

A $200/m max subscriber using OpenCode and not wanting to use API keys with pay-per-token pricing is very clearly trying to get around paying for tokens.

Is there any limits to that users 200/month? Why should they not be able to use the limits to the extent from other tools?

If openclaw chews my 200/month up in 15 days... I don't get more requests for free

There is no monthly limit, it (currently) is a weekly and 5-hourly limit. If they allow anyone to use any tool with their subscription service, you could have a system (like OpenClaw) which involves 0 human interaction and is constantly consuming 100% of your token limit, then waiting until limits reset to do it all over again. It seems fairly clear that Anthropic is probably losing money on such usage patterns.

Once again: you can use API keys and pricing to get UNLIMITED usage whenever you want. If you are choosing to pay for a subscription instead, it is because Anthropic is offering those subscriptions at a much better value-per-token. They are not offering such a subscription out of the goodness of their heart.

There are 4 weeks in a month.

4 periods of weekly limits, is a monthly limit.

4 weeks * 12 months = 48 weeks in a year * 7 days in a week = 336 days per year - close enough :)

I wonder if it has to do with Grok somehow. They had a suspiciously high reputation until they just binarily didn't, after Anthropic said they did something.

For sure, yes. They already added attempts to block opencode, etc.

the subscription already has usage caps. if the caps are the caps, why does the client matter. if the caps aren't actually the caps, that's a different conversation.

The analogy I like to use when people say "I paid" is that you can't pay for a buffet then get all the food take-home for free.

Why does it matter to Anthropic if my $200 plan usage is coming from Claude Code or a third party?

Doesn’t both count towards my usage limits the same?

If you buy a 'Season's Pass' for Disneyland, you cant 'sublet' it to another kid to use on the days you don't; It's not really buying a 'daily access rate'.

Anthropic subs are not 'bulk tokens'.

It's not an unreasonable policy and it's entirely inevitable that they have to restrict.

I’m not subletting my sub to anyone. I’m the only one using the third party harness.

I’m using their own SDK in my own CLI tool.

It’s not a literal sublet to someone else, it’s subletting your tokens to another tool.

At its core it’s a tragedy of commons situation. Using a third party tool like OpenClaw is augmenting your usage far beyond what was anticipated when the subscription plan was made.

Same deal for unlimited storage on drive until people started abusing it.

My Claude sub isn’t unlimited.

I didn’t set the limits on the plan; change those if it’s a problem, not irritate your customer base.

You have strong dedication towards taking things literally.

The issue is not that it's limited or unlimited, but rather about expected token usage across a user cohort. When you set a usage limit on something like Claude, or a gym, or a tutoring center, you need to do two things at once; set the limit high enough to attract the aspirations of your intended client base ("oh good this gym lets me go every day of the month if I want to"), but priced accurately enough so that you actually turn a profit on the average usage across most users (you ended up going 20 times the first month, but settled into 15 times a month after).

If there was suddenly a drug that you could take that would, while you slept, make your body walk to the gym and workout, so that you could max out that usage, the gym would be entitled to adjust either the pricing, the limit, or prohibit going to the gym while on the drug, given that they can't actually sustain all their members going every day.

As a correction, I've done some reading and when I said tragedy of the commons, what would fit better is a "congestion externality in a club good".

Running with the Disney analogy, it's like if Disney didn't let you wear a shirt with a universal or Warner property on it in their parks

Absurd, and not beyond the realm of possibility

It’s still me going to Disneyland, I just take a different route

Disingenuous analogy.

It's more buying a season pass for Disneyland, then getting told you can't park for free if you're entering the park even though free parking is included with the pass. Still not unreasonable, but brings to light the intention of the tool is to force the user into an ecosystem rather.

It's not a disingenuous analogy ... whatever it is.

But 'you can't park even though the ticket includes parking' is not an appropriate analogy because 3rd party use is definitely not intended. They did not 'state one thing' and the 'disallow it'.

This is a pretty straight forward case of people using their subscription for 'adjacent' use, and Anthropic being more explicit about it.

There's nothing fancy going on here.

Disingenuous or not, it was a bad analogy because it inferred that it was intentionally being abused which is completely false. The proof of that is this original post - Anthropic did not clearly (or even at all) identify how you could use your tokens with the subscription regardless of their intentions.

You're now misinterpreting my argument and misrepresenting it. I did not, in any way, suggest that Anthropic was "pulling the rug" to its users nor that they were entitled to use their tokens using the API with third parties. Full stop.

Of course, third-party API usage wasn't intended to be allowed for consuming subscription tokens. This is exactly what my analogy was structured to explain; a Disneyland season pass isn't intended to be used solely for parking. Anthropic did not intend for subscription tokens to be consumed by third-parties the same way users did not intend to abuse the subscription to derive more value than what was allotted to them. Your analogy missed that last part, which is absolutely crucial to understand.

I don't understand how you're making the exact arguments I'm making, then somehow completely misunderstanding what's being said.

They don't get as much visibility into your data, just the actual call to/from the api. There's so much more value to them in that, since you're basically running the reinforcement learning training for them.

Increasing the friction of switching providers as much as possible is part of their strategy to push users to higher subscription tiers and deny even scraps to their competitors.

Probably because the $20 plan is essentially a paid demo for the higher plans.

They're losing money on this $200 plan and they're essentially paying you to make you dependent on Claude Code so they can exploit this (somehow) in the future.

It's a bizarre plan because nobody is 'dependent' on Claude Code; we're begging to use alternatives. It's the model we want!

And it's their 'ecosystem' they want to sell you.

You’re not really paying for the model, you’re paying for the tool, the ecosystem, and the application layer around it.

Sonnet 4.6 in CC doesn’t behave the same way as Sonnet 4.6 in Antigravity.

Skills is a generic construct. System prompt is generic as well. Subagents, AGENTS.md, CLAUDE.md etc. these are generic, "please care for my instruction" kind of constructs without any real guarantee to close gaps.

Tool is generic (CC vs OpenCode) Ecosystem is already same everywhere.

I don't understand what's the point.

The point is that wrappers matter. Orchestration, tool calls, reasoning loops, system prompts, agentic capabilities. Output is different, quality is different.

This is the moat for AI frontier companies.

When using Claude Code, it's possible to opt out of having one's sessions be used for training. But is that opt out for everything? Or only message content, such that there could remain sufficient metadata to derive useful insight from?

Any user who is using a third-party client is likely self-selected into being a power user who is less profitable.

Not sure what the problem is, I am on Max and use Claude Code, never get usage issues, that's what I pay for and want that to always be an option (capped monthly cost). For other uses it makes sense to go through their API service. This is less confusing and provides clarity for users, if you are a first party user use Claude's tools to access's the models otherwise API

What is the point of developing against the Agent SDK after this change.

how can they even enforce this? can't you just spoof all your network requests to appear like it's coming from claude code?

in any case Codex is a better SOTA anyways and they let you do this. and if you aren't interested in the best models, Mistral lets you use both Vibe and their API through your vibe subscription api key which is incredible.

> how can they even enforce this?

Many ways, and they’re under no obligation to play fair and tell you which way they’re using at any given time. They’ve said what the rules are, they’ve said they’ll ban you if they catch you.

So let’s say they enforce it by adding an extra nonstandard challenge-response handshake at the beginning of the exchange, which generates a token which they’ll expect on all requests going forward. You decompile the minified JS code, figure out the protocol, try it from your own code but accidentally mess up a small detail (you didn’t realize the nonce has a special suffix). Detected. Banned.

You’ll need a new credit card to open a new account and try again. Better get the protocol right on the first try this time, because debugging is going to get expensive.

Let’s say you get frustrated and post on Twitter about what you know so far. If you share info, they’ll probably see it eventually and change their method. They’ll probably change it once a month anyway and see who they catch that way (and presumably add a minimum Claude Code version needed to reach their servers).

They’ve got hundreds of super smart coders and one of the most powerful AI models, they can do this all day.

the internet has hundreds of thousands of super smart coders with the most powerful ai models as well, I think it's a bit harder than you're assuming.

you just need to inspect the network traffic with Claude code and mimic that

OK sure, but you’d better hope Claude Code gets it right on the first try, or that’s $200 down the drain. Also, what if the detection mechanism involves a challenge-response that happens once a week? Or randomly a couple times a month? Or after 15 minutes of use from a new public IP? Or arbitrarily if you ask it to code something with a particular topic?

There are lots of ways they could be doing this. And remember again, if they get you, they don’t have to tell you how they got you (so you might not be able to even glean information in return for the $200 you’d be losing).

Sure the internet has hundreds of thousands of super smart coders, but the subset who are willing to throw money and credit cards down the drain in order to maintain a circumvention strategy for something like this is pretty low. I’m sure a few people will figure it out, but they won’t want to tell anyone lest Anthropic nerf their workaround, so I doubt that exploits of this will become widespread.

And if you’re Anthropic, that’s probably good enough.

Then just run Claude Code in a PTY and proxy requests in and out?

> Then just run Claude Code in a PTY and proxy requests in and out?

Exactly something I said too. There are projects which can do this and hook natively to opencode and even its sdk/api.

https://news.ycombinator.com/item?id=47069299#47070204 (I list a project which does this)

I really don't know how anthropic can somehow detect something like this.

When they blocked OpenCode, I was in the middle of adding a feature. I don't think it's possible to mimic CC in an undetectable way and have the feature work.

The feature allows the LLM to edit the context. For example, you can "compact" just portions of the conversation and replace it with a summary. Anthropic can see that the conversation suddenly doesn't share the same history as previous API calls.

In fact, I ported the feature to Claude Code using tweakcc, so it literally _is_ Claude Code. After a couple days they started blocking that with the same message that they send when they block third party tools.

If your service's traffic is literally indistinguishable from Claude Code, then all it can do is what Claude Code does. Then why are the users going to choose your service instead of Claude Code?

Maybe because the UI doesn't flicker? There's a lot you can do to improve the UI for starters, and then the harness around the agent could also be improved upon, as long as the prompts are the same.

easily "bypassable", trust me :)

see my comment here but I think instead of worrying about the decompile minified JS code etc., you can just essentially use claude code in the background and still do it even using opencode/its SDK thus giving sort of API access over CC subscription https://news.ycombinator.com/item?id=47069299#47070204

I am not sure how they can detect this. I can be wrong, I usually am but I think its still possible to use CC etc. even after this change if you really wanted to

But at this point, to me the question of GP that is that is it even worth it is definitely what I am thinking?

I think not. There are better options out there, they mentioned mistral and codex and I think kimi also supports maybe GLM/z.ai as well

Pretty easy to enforce it - rather than make raw queries to the LLM Claude Code can proxy through Anthropic's servers. The server can then enforce query patterns, system prompts and other stuff that outside apps cannot override.

And once all the Claude subscribers move over to Codex subscriptions, I'd bet a large sum that OpenAI will make their own ToS update preventing automated/scripted usage.

They can't catch everything but they can make your product you're building on top of it non viable when it gets popular enough to look for, like they did with opencode.

at least with open code you can just use a third party plugin to authenticate

> how can they even enforce this?

I would think that different tools would probably have different templates for their prompts?

You could tell by the prompt being used.

We don’t enforce speed limits, but it sucks when you get caught.

OpenAI will adjust, their investors will not allow money to be lost on ”being nice” forever, not until they’re handsomely paid back at least.

I'm wondering: why now, in early 2026? Why not last year? Why not in July? What changed? What does this teach us about Anthropic and what can we infer about their competition?

Especially how generous Anthropic has been recently with subscribers - extra usage in December, $50 credit earlier this month, $20 subscription getting access to Opus.

It suggests to me Anthropic is less concerned with the financial impact of letting subscribers use alternative tools and more concerned with creating lock in to their products and subscriptions. It very well might backfire though, I was not considering alternative models yesterday, but today I am actively exploring other options and considering cancelling my sub. I've been using my subscription primarily through pi recently, so if they aren't interested in me as a customer, pretty much everyone else is.

Product usage subsidized by company, $100. Users inevitably figure out how to steal those subsidies, agents go brrrrr. Users mad that subsidy stealing gets cut off and completely ignore why they need to rely on subsidies in the first place, priceless.

This article is somewhat reassuring to me, someone experimenting with openclaw on a Max subscription. But idk anything about the blog so would love to hear thoughts.

https://thenewstack.io/anthropic-agent-sdk-confusion/

In my opinion (which means nothing). If you are using your own hardware and not profiting directly from Claude’s use (as in building a service powered by your subscription). I don’t see how this is a problem. I am by no means blowing through my usage (usually <50% weekly with max x5).

¡Quick reminder! We are in the golden era of big company programming agents. Enjoy it while you can because it is likely going to get worse over time. Hopefully, there were will be competitive open source agents and some benevolent nerds put together a reasonable service. Otherwise I can see companies investing in their own AI infrastructure and developers who build their own systems becoming the top performers.

This is the VC funded startup playbook. It has been repeated many times, but maybe for the younger crowd it is new. Start a new service that is relatively permissive, then gradually restrict APIs and permissions. Finally, start throwing in ads and/or making it more expensive to use. Part of the reason is in the beginning they are trying to get as many users as possible and burning VC money. Then once the honey moon is over, they need to make a profit so they cut back on services, nerf stuff, increase prices and start adding ads.

Hot take: trying to restrict what front end people use to access your service is almost always an anti-competitive, anti-consumer freedom move which should be legally prohibited for those reasons. (Not just for AI, I'm talking about any and all cloud services.)

Regarding consumer freedom, I believe software running on user machines should serve the interests of the user, not the company who wrote the software or anyone else for that matter. Trying to force users to run a particular client written by your company violates this principle.

Regarding competition, forcing users to run a particular client is a form of anti competitive bundling, a naked attempt to prevent alternative clients from being able to enter the market unless they are able to build a competing backed as well. Such artificial "moats" are great for companies but harmful to consumers.

Anthropic has no authority to do as such. Users and third apps are protected by interoperability exceptions found in copyright case law.

Trying to prevent competitors from interoperating with the service also may be construed as anticompetitive behaviour.

The implementation details of an authentication process do not beget legal privileges to be a monopolist. What an absurd thought.

The people who they’re going to piss off the most with this are the exact people who are the least susceptible to their walled garden play. If you’re using OpenCode, you’re not going to stop using it because Anthropic tells you to; you’re just going to think ‘fuck Anthropic’, press whatever you’ve bound “switch model” to, and just continue using OpenCode. I think most power users have realized by now that Claude Code is sub-par software and probably actively holding back the models because Anthropic thinks they can’t work right without 20,000 tokens worth of system prompt (my own system prompt has around 1,000 and outperforms CC at every test I throw it at).

They’re losing the exact crowd that they want in their corner because it’s the crowd that’s far more likely to be making the decisions when companies start pivoting their workflows en-masse. Keep pissing on them and they’ll remember the wet when the time comes to decide whom to give a share from the potentially massive company’s potentially massive coffers.

Does this mean that in an absurd way you can get banned if you use CodexBar https://github.com/steipete/CodexBar to keep track of your usage? It does use your credentials to fetch the usage, could they be so extreme that this would be an issue?

I think that their main problem is that they don't have enough resources to serve too many users, so they resort to this kind of limitations to keep Claude usage under control. Otherwise I wouldn't be able to explain a commercial move that limits their offer so strongly in comparison to competitors.

That's it. That's all the moat they have.

I wrote a mcp bridge so that I don't have to copy and paste prompt back and forth between CLI and claude, chatgpt, grok, gemini

https://github.com/agentify-sh/desktop

Does this mean I have to remove claude now and go back to copy & pasting prompts for a subscription I am paying for ?!

wth happened to fair use ?

OpenAI has endorsed OAuth from 3rd party harnesses, and their limits are way higher. Use better tools (OpenCode, pi) with an arguably better model (xhigh reasoning) for longer …

I am looking forward to switching to OpenAI once my claude max account is banned for using pi....

Am I the only one perplexed why folks find this stunning or meaningful. While LLMs are novel and different in that the subscription gives you access to compute, it does not feel foreign from the subscription, free or paid, landscape. I cannot recall many (if any?) companies that would freely let you use compute, private internal APIs or anything similar just because you have a login. Maybe I come from a different era of tech but it seems both reasonable and not surprising.

Why now? It would not surprise me that this was simply an after thought and once it hit critical mass (opencode) they locked it down.

Seems fair enough really, not that I like it either, but they could easily not offer the plans and only have API pricing. Makes it make more sense to have the plans be 'the Claude Code pricing' really.

Sounds like a panicking company grasping and clawing for a moat

Their model actually doesn't have that much of a moat if at all. Their agent harness also doesn't, at least not for long. Writing an agent harness isn't that difficult. They are desperately trying to stay in power. I don´t like being a customer of this company and am investing lots of my time in moving away from them completely.

They are obviously losing money on these plans, just like all of the other companies in the space.

They are all desperately trying to stay in power, and this policy change (or clarification) is a fart in the wind in the grand scheme of what's going on in this industry.

This month was the first month i spent >$100 on it and it didn't feel like it was money well spent. I feel borderline scammed.

I'm just going to accept that my €15 (which with vat becomes €21) is just enough usage to automate some boring tasks.

Not surprised, its the official stance by Anthropic.

I'm more surprised by people using subscription auth for OpenClaw when its officially not allowed.

In the old days, think Gmail, or before the "unlimited" marketing scam. People genuinely are smart enough to know they are doing something that they are not suppose to be doing. Even Pirating software, say Windows or Adobe. I mean who can afford those when they were young?

Things get banned, but that is OK along as they give us weeks or days to prep for alternative solution. Users ( Not Customers ) are happy with it. Too bad, the good days are over.

Somewhere along the line, no just in software but even in politics, the whole world on entitlement. They somehow believe they deserve this, what they were doing were wrong but if it is allowed in the first place they should remain allowed to do so.

Judging from account opening time and comments we can also tell the age group and which camp they are on.

I don’t understand which camp are you on?

The one you're not in

Who pissed in your coffee?

It's a bit unclear to me. I'm building a system around the Claude Agent SDK. Am I allowed to use it or not? Apparently not.

At this point, are there decent alternatives to Anthropic models for coding that allow third-party usage?

OpenAI have been very generous in their plans in terms of token and what you use it with. Is Codex better or as good as Opus for coding? No. Is it a decent alternative? Very.

Thanks for the reply. Need to try Codex

Kimi is amazing for this. They offer API usage as well iirc if you buy their subscription.

Not regular api usage, just the kimi coding plan, which you can only use in some coding agents

Thanks, will explore Kimi. Haven’t tried it yet

Also the .99c deal has API Access

Yes!

Honestly seeing throttling of AI usuage across all providers:

- Google reduced AI Studio's free rate limits by 1/10th

- Perplexity imposing rate limits, card filing to continue free subscriptions

- Now Anthropic as well

There has been a false narrative that AI will get cheaper and more ubiquitous, but model providers have been stuck in a race for ever more capabilities and performance at higher costs.

I'm a bit lost on this one.

I can get a ridiculous amount of tokens in and out of something like gpt-5.2 via the API for $100.

Is this primarily about gas town and friends?

Isn’t this flawed anyway? If an application communicates with Claude Code over ACP (like Zed), it works fine?

Instead of using SDKs, this will just shift the third party clients to use ACP to get around it - Claude Code is still under the hood but you’re using a different interface.

This all seems pretty idiotic on their part - I know why they’re trying it but it won’t work. There will always be someone working around it.

What about using claude -p as an api interface?

Too bad will stick with codex as thinker and glm5 as hands, at a fraction of the cost.

This confirms they're selling those subscriptions at a loss which is simply not sustainable.

They probably are but I don’t think that’s what this confirms. Most consumer flat rate priced services restrict usage outside of the first party apps, because 3rd party and scripted users can generate orders of magnitude more usage than a single user using the app can.

So it makes sense to offer simple flat pricing for first party apps, and usage priced apis for other usage. It’s like the difference between Google Drive and S3.

I get your point - they might count on the user not using their full quota they're officially allowed to use (and if that's the case, Anthropic is not losing money). But then still - IF the user used the whole quota, Anthropic loses.. so what's advertised is not actually honest.

For me, flat rates are simply unfair either ways - if I'm not using the product much, I'm overpaying (and they're ok with that), otherwise it magically turns out that it's no longer ok when I actually want to utilize what I paid for :)

At any rate, they offer the option to be billed exactly on your usage via the API. But if you tell the average person the service costs $x/million tokens they will have no idea how much that actually costs, they won't know what a token is or how much their employees are likely to use. While $30/user/month is something they can easily budget for and get approved.

My alt Google accounts were all banned from Gemini access. Luckily Google left my main account alone. They are all cracking down.

From 3rd party AI app use?

Using a proxy to switch accounts

The reason I find this so egregious is because I don’t want to use Claude Code! It’s complete rubbish, completely sidelines security, and nobody seems to care. So I’m forced to use their slop if I want to use Claude models without getting a wallet emptying API bill? Forget it, I will use Codex or Gemini.

Claude Code is not the apex. We’re still collectively figuring out the best way to use models in software, this TOS change kills innovation.

So even simple apps that are just code usage monitors are banned?

Always have been, unless you're using the API meant for apps.

But if you're doing something very basic, you might be able to slop together a tool that does local inferencing based on a small, local model instead, alleviating the need to call Claude entirely.

Why do I get the nagging suspicion their 1 million LOC codebase is backdoored?

You need a company with a market cap in the trillions to succeed here

How does this impact open router?

Can’t this restriction for the time being be bypassed via -p command line flag?

OpenRouter uses the API and does not use any subscription auth.

Cancelled my Claude and bought GLM coding plan + Codex.

This is something I think Anthropic does not get. They want to be Microsoft of AI, make people their solution, so they will not to move to the other provided. Thing is, giving access to a text prompt is not something that you can monopolize easily. Even if you provide some stuff like skills, MCP server integration, that is not a big deal.