CursorBench 3.1: leaderboard
Cursor coding-agent benchmark for ambiguous, multi-file real-world tasks, including codebase understanding, bug finding, planning, and code review.
Metric: Score (%). Source: cursor.com. Status: saturation imminent. 49 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5.1 (Max) | 73.4 |
| 2 | Claude Fable 5.1 (xHigh) | 72.8 |
| 3 | Grok 4.6 (xHigh) | 70.8 |
| 4 | Claude Fable 5 (Max) | 70.5 |
| 5 | Claude Opus 5 (Max) | 70 |
| 6 | Grok 4.6 (High) | 69.9 |
| 7 | Claude Fable 5.1 (High) | 69.4 |
| 8 | Claude Opus 5 (xHigh) | 69.3 |
| 9 | Gemini 3.8 Flash (High) | 69.2 |
| 10 | GPT-5.6 Sol (Max) | 67.2 |
| 11 | Grok 4.6 (Medium) | 67.1 |
| 12 | Gemini 3.8 Flash (Medium) | 67 |
| 13 | Claude Opus 5 (High) | 66.7 |
| 14 | Claude Fable 5 (High) | 66.5 |
| 15 | Claude Fable 5 (Medium) | 65.2 |
Interactive version: theaggregate.ai/benchmark?slug=cursorbench-3-1 · How It Works · Data refreshed daily, snapshot 2026-09-05.