Calibration
ProgramBench → Vals LiveCodeBench
Vals LiveCodeBench is estimated from ProgramBench with a offset logistic curve fitted on 6 models measured on both: y = 0.7737 + (0.8975 − 0.7737) / (1 + exp(−144.06·(x − 0.7686))), R² = 0.68, cross-validated error 6.1 pp. It is used for 3 estimates.
| Estimated model | ProgramBench | Vals LiveCodeBench | Source |
|---|---|---|---|
| DeepSeek V4.1 Flash | 20.3% | 77.4% | estimated ± 6.1 pp, low confidence |
| MiMo-V2.6-Flash | 26.0% | 77.4% | estimated ± 6.1 pp, low confidence |
| MiMo-V2.6-Pro | 26.5% | 77.4% | estimated ± 6.1 pp, low confidence |