Cursor AI says 85% of Composer 2.5 compute goes to its own custom training and RL
Moonshot's Kimi base models accounted for the remaining 15%.
——0——
Moonshot's Kimi base models accounted for the remaining 15%.
Users criticized Cursor's Composer 2.5 as unusable and inferior to GPT despite the high custom training compute usage.
2 comments with sentiment.