Cursor launches Composer 2 AI coding model, based on Kimi K2.5, with major benchmark gains

Cursor has launched Composer 2, its latest AI coding model, expanding the company's product offerings. According to the company, Composer 2 delivers substantial improvements on established benchmarks such as Terminal-Bench 2.01 and SWE-bench Multilingual. These enhancements result from Cursor's first continued pretraining run, followed by reinforcement learning focused on long-horizon coding tasks that require hundreds of sequential actions.

Composer 2 is available at a rate of $0.50 per million input tokens and $2.50 per million output tokens. To address varied user needs, Cursor has introduced a faster variant delivering the same intelligence at $1.50 per million input and $7.50 per million output tokens, which is now set as the default. The company notes this fast option is less expensive than other comparable models.

Composer 2 usage for individual plan subscribers is handled within a standalone pool that includes substantial usage as standard. As highlighted by discussions on X, Composer 2 is based on the Kimi K2.5 AI model, providing transparency regarding its underlying technology.