Cursor Launches Composer 2 — Its Own Frontier Coding Model, Beats Claude Opus 4.6

Cursor Blog·high signal

Cursor released Composer 2, a proprietary coding model trained via reinforcement learning on long-horizon coding tasks, scoring 61.3 on CursorBench vs. Claude Opus 4.6's 58.2. Priced at $0.50/M input and $2.50/M output tokens — a fraction of frontier model costs — it supports 200K context and direct CLI interaction. Cursor now competes directly with Anthropic and OpenAI rather than relying on their models.