Calibration
CursorBench 3.2 → VulcanBench v3
VulcanBench v3 is estimated from CursorBench 3.2 with a Michaelis–Menten + offset curve fitted on 8 models measured on both: y = 0.0000 + 2.0000·x / (0.88337 + x), R² = 0.44, cross-validated error 5.1 pp. It is used for 6 estimates.
| Estimated model | CursorBench 3.2 | VulcanBench v3 | Source |
|---|---|---|---|
| Claude Opus 4.8 | 62.3% | 82.7% | estimated ± 5.1 pp, low confidence |
| Gemini 3.5 Flash | 48.8% | 71.2% | estimated ± 5.1 pp, low confidence |
| Gemini 3.6 Flash | 53.5% | 75.4% | estimated ± 5.1 pp, low confidence |
| GLM-5.2 | 55.0% | 76.7% | estimated ± 5.1 pp, low confidence |
| GPT-5.5 | 58.4% | 79.6% | estimated ± 5.1 pp, low confidence |
| Kimi K2.7 Code | 49.7% | 72.0% | estimated ± 5.1 pp, low confidence |