Calibration
Vals MMLU-Pro → HLE w/o tools
HLE w/o tools is estimated from Vals MMLU-Pro with a offset logistic curve fitted on 19 models measured on both: y = 0.2510 + (0.5738 − 0.2510) / (1 + exp(−64.54·(x − 0.8764))), R² = 0.86, cross-validated error 5.1 pp. It is used for 9 estimates.
| Estimated model | Vals MMLU-Pro | HLE w/o tools | Source |
|---|---|---|---|
| Gemini 3.1 Flash-Lite | 86.2% | 34.2% | estimated ± 5.1 pp, medium confidence |
| GLM-5.3-Flash | 86.1% | 33.8% | estimated ± 5.1 pp, medium confidence |
| Qwen3.5 Flash | 84.1% | 28.1% | estimated ± 5.1 pp, medium confidence |
| MiMo-V2.5 | 82.9% | 26.5% | estimated ± 5.1 pp, medium confidence |
| GLM-4.5 | 81.2% | 25.6% | estimated ± 5.1 pp, medium confidence |
| Claude Haiku 4.5 | 78.7% | 25.2% | estimated ± 5.1 pp, medium confidence |
| Mistral Medium 3.5 128B | 75.3% | 25.1% | estimated ± 5.1 pp, low confidence |
| Laguna XS.2 | 69.1% | 25.1% | estimated ± 5.1 pp, low confidence |
| Laguna M.1 | 68.8% | 25.1% | estimated ± 5.1 pp, low confidence |
Fitted on
| Model | Vals MMLU-Pro | HLE w/o tools |
|---|---|---|
| Claude Fable 5.1 | 92.4% | 60.9% |
| Claude Opus 4.8 | 89.6% | 49.8% |
| Claude Opus 5 | 91.6% | 56.3% |
| Claude Sonnet 5 | 87.5% | 43.2% |
| Gemini 3.1 Pro | 91.0% | 45.4% |
| GLM-5.2 | 86.7% | 40.5% |
| GPT-5.4 mini | 84.6% | 28.2% |
| GPT-5.4 nano | 77.2% | 24.3% |
| GPT-5.5 | 88.1% | 41.4% |
| Grok 4.20 | 86.3% | 31.6% |
| Inkling | 86.3% | 30.0% |
| Inkling-Small | 85.6% | 31.6% |
| Kimi K3 | 88.0% | 43.5% |
| MiMo-V2.5-Pro | 84.6% | 34.0% |
| Muse Spark | 87.3% | 42.8% |
| Muse Spark 1.1 | 88.7% | 52.2% |
| Nemotron 3 Ultra | 85.8% | 26.7% |
| Qwen3.8-27B | 84.3% | 30.8% |
| Qwen3.8 Max | 88.6% | 43.6% |