Cursor's proprietary coding model powering Composer agent mode. Tops CursorBench (61.3), Terminal-Bench (61.7), and SWE-bench Multilingual (73.7). Trained specifically for multi-file editing and agentic coding workflows.
| 信号 | 强度 | 权重 | 影响 |
|---|---|---|---|
| Benchmarksjust now | 66 | 30% | +19.7 |
| Recencyjust now | 100 | 15% | +15.0 |
| Pricingjust now | 98 | 15% | +14.6 |
| Capabilitiesjust now | 50 | 20% | +10.0 |
| Context Windowjust now | 84 | 10% | +8.4 |
| Output Capacityjust now | 80 | 10% | +8.0 |
把当前模型放回同一服务商最近的发布节奏中查看。
社区和从业者反馈在基准测试和价格之上增加了真实世界的信号。
Share your experience with Composer 2 and help the community make better decisions.
成本估算器
每月比类别平均节省$34.64