I've downgraded my Anthropic subscription back to Pro. Anthropic's walled garden keeps me out - not in.

Important disclaimer: I absolutely understand why Anthropic closed the loop hole. My issue is with how they did it.

By now you probably know what the crackdown was about: Anthropic closing a loop hole where 3rd party tools spoofed their OAuth headers, impersonating as official Anthropic tools.

Anthropic's usage limits are enforced at the application layer, and not in their backend. So as Anthropic servers are not enforcing usage limits, and the spoofing clients of course didn't enforce them, users were able to burn token volume that Hacker News users estimated would cost $1,000 or more on a straight API plan.

I do not condone that, and I don't have an issue with Anthropic blocking that kind of usage.

I do, however, have an issue with Anthropic blocking me from using their models within the limits set by my plan unless I stay within their harness.

I want the freedom to use whatever tools I choose, and then pay for the models I think are best for a given use case.

That is what I did before the crackdown (I was using Sonnet and Opus in a way that was within the limits of my Max plan), and if Anthropic would just allow me to use my Max subscription from OpenCode and other clients, and enforce my usage limits server-side, I'd be happily paying for Max going forward.

Now, you might say: "But you can still do that, Marius! That's what the Claude API is for!", and you would be right. However, pay-per-use with the API prices is not at all comparable to the subscription pricing (I encourage you to check out this amazing deep dive into Claude pricing and limits by a young mysterious Internet person 🙂). Now, differentiating on prices like that is, of course, Anthropic's prerogative. It just means that instead of me continuously increasing my Claude usage and probably eventually upgrading again down the line from 5x to 20x, I've now downgraded back to Pro, changing my workflow to using OpenRouter more diligently to switch between models and providers.

Any time a company tightens its systems and tries to wall us in, that's when we should take any steps necessary to break free.

Nothing is keeping me from using other providers. I chose Anthropic because I enjoy the quality of their models, because I supported their opposition to the DOD and because they seemed to care at least somewhat about the difficult moral questions that come with the introduction of AI in society.

But I am not dependent on being within their ecosystem. My agent memory is safely stored outside of Anthropic's systems, and so whatever agent I call has the same context to work from. So, from today, Anthropic will no longer be my go-to models. They will just be one provider in a pool of many.

As always, the way to pull users in is to provide a kickass product your users want to use. Not building walls around them that are hard to escape.