Comment by arcanemachiner
17 days ago
I believe they have some very good training data because of all the data generated by people using the service.
This is the same data they used to finetune Kimi K2.5 to make their newer Composer models, which benchmark substantially better than Kimi K2.5.
I've heard they also want to build their own base models, which will also benefit from their large amount of high-quality training data. Which will solve Grok's model quality problem.
This is all unsourced conjecture of course. But it's what I've heard.
Also from what I understand (not my day job) we're now at the point where the post-training tuning (RLHF etc.) is increasingly important since pre training no longer scales.
So it's not really fair to call it "fine tuning", it's an important part of building a coding model in 2026, and cursor have done a pretty good job with Composer