Posted by rbanffy 1 day ago
There's also a request and a PR to add such option but it was closed due to "not adhering to community standards"
Reading through his X comments and GitHub comments he is behaving immaturely. I don't trust what he's saying here. Ripping out Claude API support was just throwing a tantrum. Weird given his age - he's old enough to be more mature.
Even as a CC user I’m glad someone is forcing the discussion.
My prediction: within two years ‘model neutrality’ will be a topic of debate. Creating lock-in through discount pricing is anti-competitive. The model provider is the ISP; the tool, the website.
That is not the point. That is a mere technicality.
You signed a contract. If you don't ignore the terms of the contract to use the product in a way that is explicitly prohibited, you're abusing the product. It is as simple as that.
They offer a separate product (API) if you don't like the terms of the contract.
Also, if you really want to get technical: the limits are under the assumption that caching works as intended, which requires control of the client. 3P clients suck at caching and increase costs. But that is not the overarching point.
> Creating lock-in through discount pricing is anti-competitive.
Literally everyone does this. OpenAI is doing this with Codex, far more than Anthropic is. It's not great but players much bigger than Anthropic are using discount pricing to create an anti-competitive advantage.
Because that could be easily resolved by factoring % cache hits into the usage limits.
> Literally everyone does this.
Never a strong justification, much as I like Anthropic in general.
Why is the 'Mercedes gas station' selling gas 85% cheaper but only to Mercedes drivers?
Why is the 'Apple electric company' selling cheaper electricity to households with Apple devices?
They're not the strongest analogies, I'll admit, but that's what it smells like to me.
Absolutely not, you are not thinking from a product perspective at all.
You might not want to capture cache % hits in usage limits because there may be some edge cases you want to support that have low hits even with an optimized client. Maybe your caching strategy isn't perfect yet, so you don't count hits to keep a good product experience going.
OSS clients that freeload on the subscription break your ability to support these use cases entirely. Now you have to count cache hits at the expense of everyone else. It is a classic case of some people ruining the experience for everyone.
> Why is the 'Apple electric company' selling cheaper electricity to households with Apple devices?
Why does Netflix not let you use your OSS hacked client of choice with your subscription?
And yet, OpenAI have publicly said they welcome OpenCode users to use their subscription package. So how are they being anti-competitive "far more" than Anthropic?
It's a PR stunt. They'll eat the costs for a bit, once they've cornered the market they'll do the same thing as Anthropic.
Curious how the context window management works in practice. With large repos, the "what files to include" problem tends to dominate — does it have a strategy beyond embedding-based retrieval, or is that the main approach here?