Calibration
IFEval → IFBench
IFBench is estimated from IFEval with a Hill curve fitted on 16 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.88210^6.00 + x^6.00), R² = 0.74, cross-validated error 9.7 pp. It is used for 6 estimates.
| Estimated model | IFEval | IFBench | Source |
|---|---|---|---|
| Agents-A1 | 94.8% | 72.8% | estimated ± 9.7 pp, low confidence |
| Celeris-1 | 80.8% | 44.6% | estimated ± 9.7 pp, medium confidence |
| LFM2.5-VL-450M | 61.2% | 12.0% | estimated ± 9.7 pp, low confidence |
| Mellum2-12B-A2.5B-Instruct | 75.8% | 34.4% | estimated ± 9.7 pp, medium confidence |
| Mellum2-12B-A2.5B-Thinking | 76.5% | 35.8% | estimated ± 9.7 pp, medium confidence |
| o3-mini | 93.9% | 71.1% | estimated ± 9.7 pp, medium confidence |