Cursor launches Composer 2 AI coding model for all users

What's new? Composer 2 is an AI coding model available to all users on Cursor with an early alpha interface; it has standard and fast pricing variants with proven code benchmarks;

· 1 min read
Cursor

Cursor has just introduced Composer 2, its latest AI coding model, now available to all users on the Cursor platform and in the early alpha of its new interface. This release targets developers and technical professionals seeking a more capable and cost-efficient AI programming assistant. Composer 2 is accessible to all users, not restricted to a limited group, and features usage as part of a standalone pool with generous allowances under individual plans.

Composer 2 presents a significant leap in performance over its predecessors, as demonstrated by benchmark results: 61.3 on CursorBench, 61.7 on Terminal-Bench 2.0, and 73.7 on SWE-bench Multilingual. These scores reflect a clear improvement compared to Composer 1.5 and Composer 1, supporting claims of better handling of challenging, long-horizon coding tasks. Technical advancements stem from continued pretraining and reinforcement learning, enabling the model to solve complex problems that require hundreds of sequential actions. Composer 2 is available in two variants:

  1. A standard model priced at $0.50/M input and $2.50/M output tokens.
  2. A faster version—offered as the new default—at $1.50/M input and $7.50/M output tokens, undercutting alternative fast models in cost.

Cursor, the company behind Composer 2, has rapidly iterated on its AI coding technology, focusing on robust benchmarking and transparent pricing. By investing in advanced training techniques and making the model widely available, Cursor aims to attract developers looking for both quality and affordability in AI code generation tools.

Source