Calibration
DAYJOB Healthcare → EBR-bench
EBR-bench is estimated from DAYJOB Healthcare with a Hill curve fitted on 11 models measured on both: y = 0.2408 + (1.1662 − 0.2408)·x^0.64 / (0.15878^0.64 + x^0.64), R² = 0.75, cross-validated error 14.8 pp. It is used for 0 estimates.
Fitted on
| Model | DAYJOB Healthcare | EBR-bench |
|---|---|---|
| Claude Opus 4.8 | 0.8% | 28.6% |
| Claude Opus 5 | 9.2% | 45.7% |
| Claude Opus 5.5 | 25.2% | 71.4% |
| Gemini 3.1 Pro | 0.0% | 14.3% |
| Gemini 4 Argon | 11.6% | 64.8% |
| Gemini 4 Carbon | 15.6% | 87.6% |
| GPT-5.5 | 0.0% | 34.3% |
| GPT-5.6 Sol | 1.2% | 44.8% |
| GPT-6.1 Sol | 10.8% | 54.3% |
| GPT-6 Astra | 11.2% | 76.2% |
| GPT-6 Sol | 2.8% | 53.3% |