Cursor has published a technical report for Composer 2, outlining a two-stage recipe of continued pretraining and large-scale reinforcement learning for agentic software engineering. The company says the model reaches 61.3 on CursorBench, 61.7 on Terminal-Bench, and 73.7 on SWE-bench Multilingual while keeping pricing at $0.50/M input and $2.50/M output tokens.
Cursor said on March 26, 2026 that real-time reinforcement learning lets it ship improved Composer 2 checkpoints every five hours. Cursor’s March 27 technical report says the model combines continued pretraining on Kimi K2.5 with large-scale RL in realistic Cursor sessions, scores 61.3 on CursorBench, and runs on an asynchronous multi-region RL stack with large sandbox fleets.
Cursor has published the Composer 2 technical report, outlining its code-focused continued pretraining, large-scale reinforcement learning pipeline, and CursorBench-led evaluation strategy. The report offers an unusually detailed first-party look at how a production coding agent is trained and measured.