CursorBench
Rank #9 of 51 · 9 models scored · top score 72.9
unknown
Strongest on discrimination, weakest on saturation headroom. Caveat: the public test set leaves contamination on the table.
Weighted composite. Each signal is computed from the source-backed leaderboard below.
Does it still separate models?
How far from ceiling / clustered at the top?
Public vs held-out; training-leak risk.
Apples-to-apples, or mixed / vendor-optimized?
How old is the benchmark?
| # | Model | Score | Evidence | Measured |
|---|---|---|---|---|
| 1 | Claude Fable 5 | 72.9 | unverified | 2026-06-09 |
| 2 | Claude Opus 4.7 | 64.8 | unverified | 2026-04-16 |
| 3 | GPT-5.5 | 64.3 | unverified | 2026-04-23 |
| 4 | Claude Opus 4.8 | 63.8 | unverified | 2026-05-28 |
| 5 | GLM-5.2 | 54.6 | unverified | 2026-06-16 |
| 6 | Gemini 3.5 Flash | 49.8 | unverified | 2026-05-19 |
| 7 | Claude Sonnet 4.6 | 49 | unverified | 2026-02-17 |
| 8 | Kimi K2.6 | 47.6 | unverified | 2026-04-20 |
| 9 | Kimi K2.5 | 31.9 | unverified | 2026-02-02 |