Calibration
GPQA → HealthBench Professional
HealthBench Professional is estimated from GPQA with a Hill curve fitted on 7 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.94581^6.00 + x^6.00), R² = 0.65, cross-validated error 4.2 pp. It is used for 15 estimates.
| Estimated model | GPQA | HealthBench Professional | Source |
|---|---|---|---|
| GPT-5.2 | 92.4% | 55.8% | estimated ± 4.2 pp, medium confidence |
| Qwen3.7 Max | 92.4% | 55.8% | estimated ± 4.2 pp, medium confidence |
| Gemini 3.5 Flash | 92.2% | 55.4% | estimated ± 4.2 pp, low confidence |
| Qwen3.8-Omni-Flash | 91.0% | 53.1% | estimated ± 4.2 pp, low confidence |
| DeepSeek V4.1 Flash 1017 | 90.9% | 52.9% | estimated ± 4.2 pp, low confidence |
| Kimi K2.6 | 90.5% | 52.1% | estimated ± 4.2 pp, low confidence |
| Qwen3.6 Plus | 90.4% | 51.9% | estimated ± 4.2 pp, low confidence |
| Qwen3.7 Plus | 90.3% | 51.7% | estimated ± 4.2 pp, low confidence |
| DeepSeek V4 Pro | 90.1% | 51.3% | estimated ± 4.2 pp, low confidence |
| Grok 4.3 | 90.1% | 51.3% | estimated ± 4.2 pp, low confidence |
| Claude Sonnet 4.6 | 89.9% | 50.9% | estimated ± 4.2 pp, low confidence |
| Interfaze Beta | 89.9% | 50.9% | estimated ± 4.2 pp, low confidence |
| Qwen3.5 397B | 88.4% | 48.0% | estimated ± 4.2 pp, low confidence |
| DeepSeek V4 Flash 0731 | 88.1% | 47.4% | estimated ± 4.2 pp, low confidence |
| Qwen3.6-27B | 87.8% | 46.8% | estimated ± 4.2 pp, low confidence |
Fitted on
| Model | GPQA | HealthBench Professional |
|---|---|---|
| GPT-5.4 | 92.8% | 48.1% |
| GPT-5.6 Luna | 92.3% | 55.7% |
| GPT-5.6 Sol | 94.6% | 60.5% |
| GPT-5.6 Terra | 92.9% | 57.7% |
| GPT-6.2 Astra | 97.0% | 66.3% |
| GPT-6.2 Sol | 96.0% | 65.9% |
| GPT-6 Astra | 96.0% | 64.7% |