Comment by NorwegianDude
2 months ago
Just to point it out, Cursor has not made any good models themselves. Composer 2 is Kimi K2.5, and they tried to pass it as their own until people noticed that the api specified it as Kimi.
2 months ago
Just to point it out, Cursor has not made any good models themselves. Composer 2 is Kimi K2.5, and they tried to pass it as their own until people noticed that the api specified it as Kimi.
Cursor has released a technical paper [1] and several blog posts [2] describing the continued pretraining and RL they do on top of Kimi K2.5.
It is true that they were not transparent about the base model that they used until the model slug was discovered by a Twitter user via the API.
[1]: https://arxiv.org/abs/2603.24477 [2]: https://cursor.com/blog/real-time-rl-for-composer
I’ve tried Kimi a bit (not much to be fair) and didn’t find it to be that great. Composer feels more „real world“ capable to me, which is not a big surprise since Cursor has all the data to tune it.
Kimi is the base, but they've done tons of finetuning on top to produce a really good completions model.