Calibration
GPQA → Vals MMLU-Pro
Vals MMLU-Pro is estimated from GPQA with a offset logistic curve fitted on 24 models measured on both: y = 0.0000 + (0.8959 − 0.0000) / (1 + exp(−22.32·(x − 0.7447))), R² = 0.86, cross-validated error 1.2 pp. It is used for 60 estimates.
| Estimated model | GPQA | Vals MMLU-Pro | Source |
|---|---|---|---|
| Claude 3.5 Sonnet | 59.4% | 3.0% | estimated ± 1.2 pp, medium confidence |
| Claude Mythos 5 | 94.1% | 88.5% | estimated ± 1.2 pp, high confidence |
| Claude Opus 4.5 | 87.0% | 84.4% | estimated ± 1.2 pp, high confidence |
| Claude Opus 4.6 | 91.3% | 87.5% | estimated ± 1.2 pp, high confidence |
| Claude Opus 4.7 (Adaptive) | 94.2% | 88.5% | estimated ± 1.2 pp, high confidence |
| Claude Sonnet 4.5 | 83.4% | 78.8% | estimated ± 1.2 pp, high confidence |
| DeepSeek V3 | 59.1% | 2.8% | estimated ± 1.2 pp, medium confidence |
| DeepSeek V4.1 Flash | 90.9% | 87.4% | estimated ± 1.2 pp, high confidence |
| Gemini 2.5 Pro | 83.0% | 78.0% | estimated ± 1.2 pp, high confidence |
| Gemma 4 12B | 78.8% | 64.9% | estimated ± 1.2 pp, medium confidence |
| Gemma 4 31B | 84.3% | 80.6% | estimated ± 1.2 pp, high confidence |
| Gemma 4 E2B | 43.4% | 0.1% | estimated ± 1.2 pp, medium confidence |
| Gemma 4 E4B | 58.6% | 2.5% | estimated ± 1.2 pp, medium confidence |
| GLM-5 | 86.0% | 83.2% | estimated ± 1.2 pp, high confidence |
| GPT-4.1 | 66.3% | 12.4% | estimated ± 1.2 pp, medium confidence |
| GPT-4.1 mini | 64.2% | 8.2% | estimated ± 1.2 pp, medium confidence |
| GPT-4.1 nano | 50.3% | 0.4% | estimated ± 1.2 pp, medium confidence |
| GPT-5.2 | 92.4% | 88.0% | estimated ± 1.2 pp, high confidence |
| GPT-5.4 | 92.8% | 88.1% | estimated ± 1.2 pp, high confidence |
| GPT-6 Astra | 96.0% | 88.9% | estimated ± 1.2 pp, medium confidence |
| Granite 4.2 30B | 66.4% | 12.7% | estimated ± 1.2 pp, medium confidence |
| Granite 4.2 3B | 54.8% | 1.1% | estimated ± 1.2 pp, medium confidence |
| Granite 4.2 8B | 64.1% | 8.1% | estimated ± 1.2 pp, medium confidence |
| Hy3 Preview | 87.2% | 84.6% | estimated ± 1.2 pp, high confidence |
| Hy4 preview | 92.3% | 87.9% | estimated ± 1.2 pp, high confidence |
| Interfaze Beta | 89.9% | 86.8% | estimated ± 1.2 pp, high confidence |
| Kimi K2.5 | 87.6% | 85.0% | estimated ± 1.2 pp, high confidence |
| Kimi K2.5 (Reasoning) | 87.6% | 85.0% | estimated ± 1.2 pp, high confidence |
| LFM2.5-230M | 25.4% | 0.0% | estimated ± 1.2 pp, medium confidence |
| LFM2.5-VL-450M | 25.7% | 0.0% | estimated ± 1.2 pp, medium confidence |
| Ling 2.6 Flash | 59.0% | 2.7% | estimated ± 1.2 pp, medium confidence |
| Ling 3.0 Flash FP8 | 84.0% | 80.0% | estimated ± 1.2 pp, high confidence |
| MAI-Thinking-1 | 84.2% | 80.4% | estimated ± 1.2 pp, high confidence |
| Mellum2-12B-A2.5B-Instruct | 40.9% | 0.0% | estimated ± 1.2 pp, medium confidence |
| Mellum2-12B-A2.5B-Thinking | 57.6% | 2.0% | estimated ± 1.2 pp, medium confidence |
| MiMo-V2-Flash | 83.7% | 79.5% | estimated ± 1.2 pp, high confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 75.6% | 50.2% | estimated ± 1.2 pp, medium confidence |
| Nemotron 3 Nano Omni 30B A3B | 72.2% | 33.7% | estimated ± 1.2 pp, medium confidence |
| o1 | 75.7% | 50.9% | estimated ± 1.2 pp, medium confidence |
| o1-pro | 79.0% | 65.7% | estimated ± 1.2 pp, medium confidence |
| o3-mini | 77.2% | 58.0% | estimated ± 1.2 pp, medium confidence |
| Ornith-1.5-35B-A3B | 89.2% | 86.4% | estimated ± 1.2 pp, high confidence |
| Ornith-1.5-397B | 92.8% | 88.1% | estimated ± 1.2 pp, high confidence |
| Ornith-1.5-9B | 86.4% | 83.7% | estimated ± 1.2 pp, high confidence |
| Qwen3 235B 2507 | 77.5% | 59.4% | estimated ± 1.2 pp, medium confidence |
| Qwen3.5-122B-A10B | 86.6% | 84.0% | estimated ± 1.2 pp, high confidence |
| Qwen3.5-27B | 85.5% | 82.5% | estimated ± 1.2 pp, high confidence |
| Qwen3.5-35B-A3B | 84.2% | 80.4% | estimated ± 1.2 pp, high confidence |
| Qwen3.5 397B | 88.4% | 85.8% | estimated ± 1.2 pp, high confidence |
| Qwen3.6-27B | 87.8% | 85.2% | estimated ± 1.2 pp, high confidence |
| Qwen3.6-35B-A3B | 86.0% | 83.2% | estimated ± 1.2 pp, high confidence |
| Qwen3.7 Plus | 90.3% | 87.0% | estimated ± 1.2 pp, high confidence |
| Qwen3.8-Flash-Next | 91.7% | 87.7% | estimated ± 1.2 pp, high confidence |
| Qwen3.8-Omni-Flash | 91.0% | 87.4% | estimated ± 1.2 pp, high confidence |
| Sakana Fugu | 95.5% | 88.8% | estimated ± 1.2 pp, medium confidence |
| Sakana Fugu-Ultra | 95.5% | 88.8% | estimated ± 1.2 pp, medium confidence |
| Soofi S 30B-A3B | 43.4% | 0.1% | estimated ± 1.2 pp, medium confidence |
| Ternary Bonsai 2 27B | 85.8% | 82.9% | estimated ± 1.2 pp, high confidence |
| ZAYA1-74B-Preview | 57.3% | 1.9% | estimated ± 1.2 pp, medium confidence |
| ZAYA1-8B | 71.0% | 28.2% | estimated ± 1.2 pp, medium confidence |