Cursor 2.0 + Composer
Cursor 2.0 introduces Composer, its first in-house agentic coding model (claimed 4x faster than peers), and a multi-agent interface.
Cursor trains its own model with codebase-wide semantic search and runs many agents in parallel via git worktrees or remote machines; adds in-app code review and a native browser tool. Most turns finish under 30 seconds (company claim).
- Date
- Wednesday, 29 October 2025
- Lab
- Anysphere (Cursor)
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Speed vs similarly intelligent models | 4x faster most turns <30 s | company |
Same day Cognition shipped SWE-1.5; both built on fast custom models. Benchmarks are Cursor's own; external evaluation limited.
Sources
This record was checked against its sources on 6 October 2026. How we check