Benchmarks · coding
CursorBench 4.0 (Cursor)
Agent evaluation on ambiguous, multi-file tasks drawn from real Cursor sessions. Maintained by Cursor. Score: CursorBench score.
18 models · 75 results · data as of 2026-10-29 · cursor.com ↗
Leaderboard
Opus 5.5 max
All results
75 rows| # | Model | Effort | CursorBench 4.0 (Cursor) |
|---|---|---|---|
| 1 | Opus 5.5 | max | 57.8% |
| 2 | Opus 5.5 | high | 56.0% |
| 3 | Opus 5.5 | xhigh | 56.0% |
| 4 | Sonnet 5.5 | max | 55.5% |
| 5 | Sonnet 5.5 | xhigh | 53.1% |
| 6 | Gemini 4 Argon | high | 52.9% |
| 7 | Gemini 4 Argon | medium | 52.8% |
| 8 | Opus 5.5 | medium | 52.5% |
| 9 | Fable 5.1 | max | 51.8% |
| 10 | Fable 5.1 | xhigh | 51.6% |
| 11 | Gemini 4 Argon | low | 50.6% |
| 12 | Fable 5.1 | high | 49.2% |
| 13 | Haiku 5.5 | max | 48.4% |
| 14 | Sonnet 5.5 | high | 47.8% |
| 15 | DeepSeek V4.1 Flash 1017 | max | 47.7% |
| 16 | Fable 5.1 | medium | 46.8% |
| 17 | Opus 5 | max | 46.6% |
| 18 | DeepSeek V4.1 Flash 1017 | medium | 46.3% |
| 19 | Grok 4.7 | xhigh | 46.3% |
| 20 | Opus 5 | xhigh | 46.1% |
| 21 | DeepSeek V4.1 Flash 1017 | high | 45.6% |
| 22 | Fable 5.1 | low | 45.1% |
| 23 | DeepSeek V4.1 Flash 1017 | low | 44.9% |
| 24 | Opus 5 | high | 44.7% |
| 25 | Haiku 5.5 | xhigh | 44.3% |
| 26 | Grok 4.7 | high | 43.9% |
| 27 | Opus 5.5 | low | 43.7% |
| 28 | Opus 5 | medium | 43.3% |
| 29 | GLM 5.3 | max | 42.6% |
| 30 | Haiku 5.5 | high | 42.3% |