Calibration
VulcanBench v3 → Bug Hunt Bench
Bug Hunt Bench is estimated from VulcanBench v3 with a Hill curve fitted on 7 models measured on both: y = 0.0771 + (1.2000 − 0.0771)·x^6.00 / (1.09727^6.00 + x^6.00), R² = 0.37, cross-validated error 8.6 pp. It is used for 1 estimate.
| Estimated model | VulcanBench v3 | Bug Hunt Bench | Source |
|---|---|---|---|
| Claude Haiku 4.5 | 76.2% | 19.0% | estimated ± 8.6 pp, low confidence |
Fitted on
| Model | VulcanBench v3 | Bug Hunt Bench |
|---|---|---|
| Claude Opus 5 | 87.0% | 27.0% |
| GLM-5.3 | 78.3% | 19.0% |
| GPT-5.6 Sol | 87.0% | 42.0% |
| Grok 4.6 | 87.0% | 27.0% |
| Kimi K3 | 73.7% | 21.0% |
| Qwen3.8-27B | 82.6% | 15.0% |
| Qwen3.8 Max | 81.2% | 25.7% |