Calibration
GPQA → HealthBench Professional
HealthBench Professional is estimated from GPQA with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.95106^6.00 + x^6.00), R² = 0.53, cross-validated error 4.8 pp. It is used for 54 estimates.
| Estimated model | GPQA | HealthBench Professional | Source |
|---|---|---|---|
| Claude 3.5 Sonnet | 59.4% | 6.7% | estimated ± 4.8 pp, low confidence |
| Claude Mythos 5 | 94.1% | 58.1% | estimated ± 4.8 pp, medium confidence |
| Claude Opus 4.5 | 87.0% | 44.3% | estimated ± 4.8 pp, low confidence |
| Claude Opus 4.6 | 91.3% | 52.7% | estimated ± 4.8 pp, low confidence |
| DeepSeek V3 | 59.1% | 6.5% | estimated ± 4.8 pp, low confidence |
| DeepSeek V4.1 Flash | 90.9% | 51.9% | estimated ± 4.8 pp, low confidence |
| Gemini 2.5 Pro | 83.0% | 36.8% | estimated ± 4.8 pp, low confidence |
| Gemma 4 12B | 78.8% | 29.3% | estimated ± 4.8 pp, low confidence |
| Gemma 4 31B | 84.3% | 39.2% | estimated ± 4.8 pp, low confidence |
| Gemma 4 E2B | 43.4% | 1.1% | estimated ± 4.8 pp, low confidence |
| Gemma 4 E4B | 58.6% | 6.2% | estimated ± 4.8 pp, low confidence |
| GLM-5 | 86.0% | 42.4% | estimated ± 4.8 pp, low confidence |
| GPT-4.1 | 66.3% | 12.4% | estimated ± 4.8 pp, low confidence |
| GPT-4.1 mini | 64.2% | 10.4% | estimated ± 4.8 pp, low confidence |
| GPT-4.1 nano | 50.3% | 2.6% | estimated ± 4.8 pp, low confidence |
| Granite 4.2 30B | 66.4% | 12.5% | estimated ± 4.8 pp, low confidence |
| Granite 4.2 3B | 54.8% | 4.2% | estimated ± 4.8 pp, low confidence |
| Granite 4.2 8B | 64.1% | 10.3% | estimated ± 4.8 pp, low confidence |
| Hy3 Preview | 87.2% | 44.7% | estimated ± 4.8 pp, low confidence |
| Hy4 preview | 92.3% | 54.6% | estimated ± 4.8 pp, medium confidence |
| Interfaze Beta | 89.9% | 50.0% | estimated ± 4.8 pp, low confidence |
| Kimi K2.5 | 87.6% | 45.5% | estimated ± 4.8 pp, low confidence |
| Kimi K2.5 (Reasoning) | 87.6% | 45.5% | estimated ± 4.8 pp, low confidence |
| LFM2.5-230M | 25.4% | 0.0% | estimated ± 4.8 pp, low confidence |
| LFM2.5-VL-450M | 25.7% | 0.0% | estimated ± 4.8 pp, low confidence |
| Ling 2.6 Flash | 59.0% | 6.5% | estimated ± 4.8 pp, low confidence |
| Ling 3.0 Flash FP8 | 84.0% | 38.6% | estimated ± 4.8 pp, low confidence |
| MAI-Thinking-1 | 84.2% | 39.0% | estimated ± 4.8 pp, low confidence |
| Mellum2-12B-A2.5B-Instruct | 40.9% | 0.8% | estimated ± 4.8 pp, low confidence |
| Mellum2-12B-A2.5B-Thinking | 57.6% | 5.6% | estimated ± 4.8 pp, low confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 75.6% | 24.1% | estimated ± 4.8 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 72.2% | 19.3% | estimated ± 4.8 pp, low confidence |
| o1 | 75.7% | 24.3% | estimated ± 4.8 pp, low confidence |
| o1-pro | 79.0% | 29.7% | estimated ± 4.8 pp, low confidence |
| o3-mini | 77.2% | 26.7% | estimated ± 4.8 pp, low confidence |
| Ornith-1.5-35B-A3B | 89.2% | 48.6% | estimated ± 4.8 pp, low confidence |
| Ornith-1.5-397B | 92.8% | 55.6% | estimated ± 4.8 pp, medium confidence |
| Ornith-1.5-9B | 86.4% | 43.2% | estimated ± 4.8 pp, low confidence |
| Qwen3 235B 2507 | 77.5% | 27.2% | estimated ± 4.8 pp, low confidence |
| Qwen3.5-122B-A10B | 86.6% | 43.6% | estimated ± 4.8 pp, low confidence |
| Qwen3.5-27B | 85.5% | 41.5% | estimated ± 4.8 pp, low confidence |
| Qwen3.5-35B-A3B | 84.2% | 39.0% | estimated ± 4.8 pp, low confidence |
| Qwen3.5 397B | 88.4% | 47.0% | estimated ± 4.8 pp, low confidence |
| Qwen3.6-27B | 87.8% | 45.9% | estimated ± 4.8 pp, low confidence |
| Qwen3.6-35B-A3B | 86.0% | 42.4% | estimated ± 4.8 pp, low confidence |
| Qwen3.7 Plus | 90.3% | 50.7% | estimated ± 4.8 pp, low confidence |
| Qwen3.8-Flash-Next | 91.7% | 53.5% | estimated ± 4.8 pp, low confidence |
| Qwen3.8-Omni-Flash | 91.0% | 52.1% | estimated ± 4.8 pp, low confidence |
| Sakana Fugu | 95.5% | 60.7% | estimated ± 4.8 pp, medium confidence |
| Sakana Fugu-Ultra | 95.5% | 60.7% | estimated ± 4.8 pp, medium confidence |
| Soofi S 30B-A3B | 43.4% | 1.1% | estimated ± 4.8 pp, low confidence |
| Ternary Bonsai 2 27B | 85.8% | 42.0% | estimated ± 4.8 pp, low confidence |
| ZAYA1-74B-Preview | 57.3% | 5.5% | estimated ± 4.8 pp, low confidence |
| ZAYA1-8B | 71.0% | 17.7% | estimated ± 4.8 pp, low confidence |