Calibration
Vals GPQA Diamond → GPQA Diamond
GPQA Diamond is estimated from Vals GPQA Diamond with a Michaelis–Menten + offset curve fitted on 23 models measured on both: y = 0.0000 + 2.0000·x / (1.08507 + x), R² = 0.67, cross-validated error 1.7 pp. It is used for 14 estimates.
| Estimated model | Vals GPQA Diamond | GPQA Diamond | Source |
|---|---|---|---|
| MiniMax M3 | 92.7% | 92.1% | estimated ± 1.7 pp, high confidence |
| Claude Opus 4.7 | 90.2% | 90.8% | estimated ± 1.7 pp, high confidence |
| Claude Sonnet 5 | 88.9% | 90.1% | estimated ± 1.7 pp, high confidence |
| GLM-5.3 | 88.1% | 89.6% | estimated ± 1.7 pp, high confidence |
| Gemini 3 Flash | 87.9% | 89.5% | estimated ± 1.7 pp, high confidence |
| GLM-5.3-Flash | 86.4% | 88.7% | estimated ± 1.7 pp, high confidence |
| Gemini 3.5 Flash-Lite | 83.8% | 87.2% | estimated ± 1.7 pp, high confidence |
| Qwen3.5 Flash | 82.8% | 86.6% | estimated ± 1.7 pp, medium confidence |
| MiMo-V2.5-Pro | 82.6% | 86.4% | estimated ± 1.7 pp, medium confidence |
| MiMo-V2.5 | 81.6% | 85.8% | estimated ± 1.7 pp, medium confidence |
| Gemini 3.1 Flash-Lite | 81.1% | 85.5% | estimated ± 1.7 pp, medium confidence |
| GLM-4.6 | 74.5% | 81.4% | estimated ± 1.7 pp, medium confidence |
| Claude Haiku 4.5 | 72.2% | 79.9% | estimated ± 1.7 pp, medium confidence |
| GLM-4.5 | 72.2% | 79.9% | estimated ± 1.7 pp, medium confidence |
Fitted on
| Model | Vals GPQA Diamond | GPQA Diamond |
|---|---|---|
| Claude Opus 4.8 | 92.4% | 93.6% |
| DeepSeek V4 Flash 0731 | 89.9% | 88.1% |
| DeepSeek V4 Pro 0813 | 92.4% | 90.1% |
| Gemini 3.1 Pro | 95.5% | 94.3% |
| Gemini 3.5 Flash | 92.7% | 92.7% |
| GLM-5.1 | 84.5% | 86.2% |
| GLM-5.2 | 85.6% | 91.2% |
| GPT-5.5 | 93.2% | 93.6% |
| GPT-5.6 Luna | 91.7% | 92.3% |
| GPT-5.6 Sol | 95.2% | 94.6% |
| GPT-5.6 Terra | 90.9% | 92.9% |
| Grok 4.20 | 88.6% | 88.5% |
| Inkling | 87.1% | 87.9% |
| Inkling-Small | 83.6% | 89.5% |
| Kimi K2.6 | 89.1% | 90.5% |
| Kimi K3 | 92.9% | 93.5% |
| Ling 3.0 Flash | 84.8% | 85.0% |
| MiniMax M2.7 | 86.6% | 87.0% |
| Muse Spark | 89.6% | 89.5% |
| Nemotron 3 Ultra | 86.1% | 87.0% |
| Qwen3.7 Max | 90.2% | 92.4% |
| Qwen3.8-27B | 88.9% | 89.2% |
| Qwen3.8 Max | 93.7% | 92.6% |