CursorBench 3.2
We evaluate agents on ambiguous, multi-file tasks from real Cursor sessions. Higher scores are better.
More about CursorBench| Model | |||||
|---|---|---|---|---|---|
| 1 | Fable 5 Max | 70.5% | $17.32 | 103,525 | 72 |
| 2 | Fable 5 Extra High | 68.4% | $11.73 | 64,971 | 56 |
| 3 | Grok 4.5 High* | 66.7% | $1.51 | 19,521 | 33 |
| 4 | Fable 5 High | 66.5% | $8.77 | 43,747 | 48 |
| 5 | Grok 4.5 Medium* | 65.4% | $1.54 | 18,914 | 34 |
| 6 | Fable 5 Medium | 65.2% | $6.80 | 30,366 | 41 |
| 7 | Grok 4.5 Low* | 63.5% | $1.22 | 15,841 | 31 |
| 8 | Opus 4.8 Max | 62.3% | $5.77 | 71,411 | 44 |
| 9 | Fable 5 Low | 62.1% | $4.46 | 18,182 | 31 |
| 10 | Sonnet 5 Max | 61.5% | $6.45 | 92,882 | 86 |
| 11 | Opus 4.8 Extra High | 59.4% | $4.50 | 51,121 | 40 |
| 12 | Sonnet 5 Extra High | 58.7% | $4.16 | 52,871 | 67 |
| 13 | GPT-5.5 High | 58.4% | $2.05 | 12,183 | 28 |
| 14 | GPT-5.5 Extra High | 58.4% | $2.85 | 17,534 | 32 |
| 15 | Opus 4.8 High | 58.0% | $3.15 | 33,548 | 33 |
| 16 | Sonnet 5 High | 56.9% | $3.19 | 39,483 | 57 |
| 17 | Opus 4.8 Medium | 56.1% | $2.81 | 28,384 | 32 |
| 18 | Composer 2.5 | 56.1% | $0.44 | 14,286 | 33 |
| 19 | GLM 5.2 Max | 55.0% | $1.76 | 35,946 | 58 |
| 20 | GPT-5.5 Medium | 53.8% | $1.51 | 8,522 | 25 |
| 21 | Opus 4.8 Low | 53.1% | $2.02 | 19,624 | 27 |
| 22 | Sonnet 5 Medium | 52.4% | $2.16 | 26,200 | 46 |
| 23 | GLM 5.2 High | 51.5% | $1.19 | 21,829 | 49 |
| 24 | Kimi K2.7 Code | 49.7% | $1.43 | 31,247 | 58 |
| 25 | Gemini 3.5 Flash | 48.8% | $2.20 | 46,702 | 77 |
| 26 | Sonnet 5 Low | 47.7% | $1.30 | 16,269 | 33 |
| 27 | GPT-5.5 Low | 46.6% | $0.98 | 5,168 | 20 |
Grok 4.5 has an advantage on CursorBench: an earlier snapshot of the Cursor codebase was unintentionally included in training. The exact score impact is unclear. That data has been removed for future models. For a rundown of third-party benchmark scores, see the Grok 4.5 launch blog.
Changelog
CursorBench 3.2
- Introduced instruction following and advanced tool use problems.
CursorBench 3.1
- Introduced problems focused on codebase understanding, bugfinding, planning, and code review.
- Improved grading criteria for some edit tasks.
CursorBench 3.0
- Initial set of tasks focused on edit, refactor, and bugfix problems.
Avg cost / task is computed by applying each model's published per-million-token pricing (input, cache read, cache write, and output) to the tokens it used on each task across the CursorBench 3.2 benchmarks, then averaging with the same task weights as the CursorBench 3.2 score. Results are subject to variance; small differences in scores may not be statistically meaningful.