Calibration
VulcanBench v3 → FrontierCode 1.1 Extended
FrontierCode 1.1 Extended is estimated from VulcanBench v3 with a offset logistic curve fitted on 5 models measured on both: y = 0.5454 + (0.6584 − 0.5454) / (1 + exp(−200.00·(x − 0.8698))), R² = 0.40, cross-validated error 3.4 pp. It is used for 1 estimate.
| Estimated model | VulcanBench v3 | FrontierCode 1.1 Extended | Source |
|---|---|---|---|
| Muse Spark 1.2 | 87.0% | 60.3% | estimated ± 3.4 pp, low confidence |
Fitted on
| Model | VulcanBench v3 | FrontierCode 1.1 Extended |
|---|---|---|
| Claude Opus 5 | 87.0% | 63.6% |
| GPT-5.6 Luna | 85.5% | 55.1% |
| GPT-5.6 Sol | 87.0% | 60.6% |
| GPT-5.6 Terra | 87.0% | 55.8% |
| Grok 4.6 | 87.0% | 61.3% |