Cursor's Composer 2 debuts
Cursor released Composer 2, a code‑specialized LLM the vendor claims matches OpenAI and Anthropic on coding tasks while undercutting them with aggressive pricing. That intensifies price/performance pressure in the developer tooling segment and could reshape benchmarking for coding assistants. (explore.n1n.ai)
Composer 2 posted 61.3 on CursorBench, 61.7 on Terminal‑Bench 2.0 and 73.7 on SWE‑bench Multilingual, versus Composer 1.5’s 44.2/47.9/65.9 respectively. (cursor.com) Cursor published per‑token pricing of $0.50 per million input tokens and $2.50 per million output tokens for the standard tier, with a “fast” variant at $1.50/$7.50 per million and the fast mode set as the default. (cursor.com) The team says Composer 2 was built by running a first continued‑pretraining pass and then reinforcement learning on long‑horizon coding tasks that can require hundreds of actions. (cursor.com) Early press comparisons report Composer 2 outscoring Anthropic’s Claude Opus 4.6 on several coding benchmarks while remaining behind OpenAI’s GPT‑5.4 on the toughest tests. (venturebeat.com) Cursor states it evaluated Composer 2 using the Harbor harness for Terminal‑Bench 2.0 with five iterations per model‑agent pair, and normalized tokens‑per‑second and output pricing to account for Anthropic’s ~15% smaller token sizing. (cursor.com) Composer 2 is available inside the Cursor product and in the early alpha of Cursor’s new interface, and Cursor says Composer usage on individual plans lives in a standalone usage pool with included allowances. (cursor.com) Industry coverage frames the release and its pricing as setting a new intelligence‑per‑dollar benchmark for coding assistants, a dynamic likely to sharpen price/performance comparisons among developer tooling vendors. (thenewstack.io)
