Cursor’s Composer 2 is now available inside Cursor, positioning itself as the editor’s latest in-house “Composer” model iteration for AI-assisted coding. In the broader open-model ecosystem, the release also lands with an interesting footnote: Moonshot notes that Kimi-k2.5 provides the foundation for what Cursor ships here, with Cursor layering on continued pretraining and high-compute RL work. Cursor also notes it accesses Kimi-k2.5 via Fireworks’ hosted RL and inference platform under an authorized commercial partnership.
Frontier-level coding, measured in Cursor’s own benchmarks
Cursor frames Composer 2 as a step-change in capability across the evaluations it tracks, including its own CursorBench along with Terminal-Bench 2.0 and SWE-bench Multilingual. The published table shows clear deltas versus the prior Composer releases:
- Composer 2: CursorBench 61.3, Terminal-Bench 2.0 61.7, SWE-bench Multilingual 73.7
- Composer 1.5: CursorBench 44.2, Terminal-Bench 2.0 47.9, SWE-bench Multilingual 65.9
- Composer 1: CursorBench 38.0, Terminal-Bench 2.0 40.0, SWE-bench Multilingual 56.9
Cursor attributes the jump to its first continued pretraining run, describing it as a stronger base for scaling reinforcement learning. From there, Composer 2 is trained on long-horizon coding tasks using reinforcement learning, with Cursor saying the model can handle tasks that require hundreds of actions.
Pricing, plus a “fast” default variant
Cursor lists Composer 2 pricing at $0.50/M input tokens and $2.50/M output tokens. Alongside it is a faster variant with the same intelligence, priced at $1.50/M input tokens and $7.50/M output tokens—and Cursor says it is making fast the default option. Additional details are available in the model docs.
On individual plans, Composer usage draws from a standalone usage pool, documented in Cursor’s pricing notes: usage pools.
For benchmark context, Cursor also includes methodology notes for Terminal-Bench 2.0, referencing the Harbor evaluation framework, the official Terminal Bench website, and the Terminal-Bench 2.0 leaderboard.
