Calibration
Vals LiveCodeBench → FrontierCode 1.1 Main
FrontierCode 1.1 Main is estimated from Vals LiveCodeBench with a Hill curve fitted on 9 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.95372^6.00 + x^6.00), R² = 0.66, cross-validated error 6.2 pp. It is used for 3 estimates.
| Estimated model | Vals LiveCodeBench | FrontierCode 1.1 Main | Source |
|---|---|---|---|
| GPT-5.1-Codex | 85.6% | 41.2% | estimated ± 6.2 pp, medium confidence |
| GPT-5.1-Codex-Max | 83.6% | 37.4% | estimated ± 6.2 pp, medium confidence |
| GLM-4.6 | 81.0% | 32.7% | estimated ± 6.2 pp, low confidence |
Fitted on
| Model | Vals LiveCodeBench | FrontierCode 1.1 Main |
|---|---|---|
| Claude Fable 5 | 89.8% | 53.5% |
| Claude Opus 4.7 | 85.1% | 38.5% |
| Claude Opus 4.8 | 87.8% | 46.5% |
| Claude Opus 5 | 89.0% | 53.4% |
| Claude Sonnet 4.6 | 82.1% | 24.3% |
| Claude Sonnet 5 | 82.4% | 42.7% |
| Gemini 3.7 Flash | 88.7% | 43.6% |
| GPT-5.4 mini | 81.5% | 27.0% |
| GPT-5.5 | 85.3% | 43.0% |