This is such a weird release. Composer 1 was cheap, fast and good enough. This is more expensive than gpt5.3, and there's no comparison apart from its previous version which doesn't really give me any confidence.
Yeah this is truly bizarre. I've been assuming that their "auto" mode basically chooses their own Composer model when you run out of API tokens, but now their homebrew model is supposedly more expensive than the mainstream offerings? How does this reflect on my Cursor subscription? Why is it more expensive in the first place when it's "just" Composer 1 + more reinforcement learning on top? How can it be more expensive to run? Or have they just decided to stop subsidizing it?
This is such a weird release. Composer 1 was cheap, fast and good enough. This is more expensive than gpt5.3, and there's no comparison apart from its previous version which doesn't really give me any confidence.
Yeah this is truly bizarre. I've been assuming that their "auto" mode basically chooses their own Composer model when you run out of API tokens, but now their homebrew model is supposedly more expensive than the mainstream offerings? How does this reflect on my Cursor subscription? Why is it more expensive in the first place when it's "just" Composer 1 + more reinforcement learning on top? How can it be more expensive to run? Or have they just decided to stop subsidizing it?
Where are the benchmarks?