Cursor Composer 2 Beats Claude Opus at a Fraction of the Cost

Cursor just shipped its own coding model. Composer 2 isn't a wrapper around Claude or GPT any more — it's a model Cursor built and trained specifically for writing code, and it's beating Opus 4.6 on coding benchmarks at a fraction of the cost. Composer 2 was trained using reinforcement learning on long-horizon coding tasks, the kind that require hundreds of sequential actions to complete. The key innovation is a technique that teaches the model to summarise its own context as it works, which means it stays coherent over much longer sessions than most models manage. But this model isn't just about quality — it's also about price. Fast mode is $1.50 per million input tokens, whilst Claude's Opus is significantly more expensive. If the benchmark results hold up in practice, that's a difficult value proposition to argue with. Cursor is now in the model-making business.