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. 38 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 (Max) | 70.5 |
| 2 | GPT-5.6 Sol (Max) | 67.2 |
| 3 | Grok 4.5 (High) | 66.7 |
| 4 | Claude Fable 5 (High) | 66.5 |
| 5 | GPT-5.6 Terra (Max) | 64.9 |
| 6 | GPT-5.6 Sol (xHigh) | 64.5 |
| 7 | GPT-5.6 Sol (High) | 63.5 |
| 8 | Claude Opus 4.8 | 62.3 |
| 9 | Claude Sonnet 5 (Max) | 61.5 |
| 10 | GPT-5.6 Luna (Max) | 61.1 |
| 11 | GPT-5.6 Sol (Medium) | 60 |
| 12 | GPT-5.6 Terra (xHigh) | 59.2 |
| 13 | Claude Sonnet 5 (xHigh) | 58.7 |
| 14 | GPT-5.5 | 58.4 |
| 15 | GPT-5.6 Luna (xHigh) | 57.7 |
Interactive version: theaggregate.ai/benchmark?slug=cursorbench-3-1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.