How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL
Sequoia Capital · 45:33 · Watch on YouTube
Cursor built Composer 2, its agentic coding model, on top of the open-source Kimi K2.5. It added large-scale continued pretraining on code, then a very large reinforcement-learning (RL) run inside Cursor's own harness...