Benchmarks / Coding
CursorBench 4.0 (Cursor)
Agent evaluation on ambiguous, multi-file tasks drawn from real Cursor sessions. Maintained by Cursor. Score: CursorBench score.
CursorBench 4.0 (Cursor)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | max | 57.8% |
| 2 | Sonnet 5.5 | max | 55.5% |
| 3 | Fable 5.1 | max | 51.8% |
| 4 | Haiku 5.5 | max | 48.4% |
| 5 | Opus 5 | max | 46.6% |
| 6 | Grok 4.7 | xhigh | 46.3% |
| 7 | GLM 5.3 | max | 42.6% |
| 8 | GPT-5.6 Sol | max | 41.7% |
| 9 | Muse Spark 1.3 | max | 41.6% |
| 10 | Grok 4.6 | xhigh | 41.4% |
| 11 | GPT-5.6 Terra | max | 41.3% |
| 12 | Gemini 3.8 Flash | high | 39.6% |
| 13 | GLM 5.3 Flash | max | 36.8% |
| 14 | GPT-5.6 Luna | max | 35.9% |
| 15 | Sonnet 5 | max | 34.1% |
| 16 | Composer 2.5 | none | 27.7% |