Composer 2

Source: Cursor Date: 2026-03-19 URL: https://cursor.com/changelog/composer-2

Summary

Cursor releases Composer 2, a new proprietary coding model positioned as “frontier-level coding performance” validated on CursorBench. Two pricing tiers: Standard at $0.50/$2.50 per million input/output tokens, Fast (default) at $1.50/$7.50 per million tokens. No changes to existing model routing for other providers.

Implications

Model integration cadence. Composer 2 is the most significant signal yet that Cursor is building a vertical model stack rather than remaining a pure integrator. Publishing CursorBench results and pricing separately from Anthropic/OpenAI/Google models signals this is a strategic model bet, not just a fine-tune. If Composer 2 performs at frontier level at $7.50/M output (vs. Claude Opus at $75/M), the cost pressure on frontier providers is real.

Pricing model evolution. Tiered Fast/Standard pricing on a proprietary model is a new pricing shape for Cursor — previously, cost was bundled into subscription or passed through from provider rates. This is Cursor becoming a model vendor, not just an IDE. The Fast default at 3x Standard cost suggests Cursor is betting users will pay for latency.

Watch: CursorBench methodology and whether it becomes an industry reference, and whether Composer 2 becomes the default for agent loops (displacing Claude/GPT defaults).