Calibration
GPQA → HLE w/o tools
HLE w/o tools is estimated from GPQA with a offset logistic curve fitted on 25 models measured on both: y = 0.0602 + (0.7411 − 0.0602) / (1 + exp(−19.09·(x − 0.9183))), R² = 0.90, cross-validated error 4.9 pp. It is used for 32 estimates.
| Estimated model | GPQA | HLE w/o tools | Source |
|---|---|---|---|
| Claude 3.5 Sonnet | 59.4% | 6.2% | estimated ± 4.9 pp, medium confidence |
| Claude Opus 4.5 | 87.0% | 25.4% | estimated ± 4.9 pp, high confidence |
| Claude Sonnet 4.5 | 83.4% | 17.4% | estimated ± 4.9 pp, high confidence |
| Claude Sonnet 4.6 | 89.9% | 33.9% | estimated ± 4.9 pp, high confidence |
| DeepSeek V3 | 59.1% | 6.2% | estimated ± 4.9 pp, medium confidence |
| Gemini 2.5 Pro | 83.0% | 16.7% | estimated ± 4.9 pp, high confidence |
| Gemma 4 E2B | 43.4% | 6.0% | estimated ± 4.9 pp, medium confidence |
| Gemma 4 E4B | 58.6% | 6.1% | estimated ± 4.9 pp, medium confidence |
| GLM-4.7 | 85.7% | 22.1% | estimated ± 4.9 pp, high confidence |
| GPT-4.1 | 66.3% | 6.5% | estimated ± 4.9 pp, medium confidence |
| GPT-4.1 mini | 64.2% | 6.4% | estimated ± 4.9 pp, medium confidence |
| GPT-4.1 nano | 50.3% | 6.0% | estimated ± 4.9 pp, medium confidence |
| GPT-5.2 | 92.4% | 41.9% | estimated ± 4.9 pp, high confidence |
| Granite 4.2 30B | 66.4% | 6.6% | estimated ± 4.9 pp, medium confidence |
| Granite 4.2 3B | 54.8% | 6.1% | estimated ± 4.9 pp, medium confidence |
| Granite 4.2 8B | 64.1% | 6.4% | estimated ± 4.9 pp, medium confidence |
| Grok 4.3 | 90.1% | 34.5% | estimated ± 4.9 pp, high confidence |
| Kimi K2.5 (Reasoning) | 87.6% | 27.0% | estimated ± 4.9 pp, high confidence |
| LFM2.5-VL-450M | 25.7% | 6.0% | estimated ± 4.9 pp, medium confidence |
| Ling 2.6 Flash | 59.0% | 6.2% | estimated ± 4.9 pp, medium confidence |
| MiMo-V2-Flash | 83.7% | 17.9% | estimated ± 4.9 pp, high confidence |
| o1 | 75.7% | 9.0% | estimated ± 4.9 pp, high confidence |
| o1-pro | 79.0% | 11.4% | estimated ± 4.9 pp, high confidence |
| o3-mini | 77.2% | 9.9% | estimated ± 4.9 pp, high confidence |
| Qwen3 235B 2507 | 77.5% | 10.2% | estimated ± 4.9 pp, high confidence |
| Qwen3.5-122B-A10B | 86.6% | 24.3% | estimated ± 4.9 pp, high confidence |
| Qwen3.5-27B | 85.5% | 21.7% | estimated ± 4.9 pp, high confidence |
| Qwen3.5-35B-A3B | 84.2% | 18.9% | estimated ± 4.9 pp, high confidence |
| Qwen3.5 397B | 88.4% | 29.3% | estimated ± 4.9 pp, high confidence |
| Qwen3.6-27B | 87.8% | 27.6% | estimated ± 4.9 pp, high confidence |
| Qwen3.6-35B-A3B | 86.0% | 22.9% | estimated ± 4.9 pp, high confidence |
| Qwen3.6 Plus | 90.4% | 35.4% | estimated ± 4.9 pp, high confidence |