Calibration
Vals MMLU-Pro → HLE
HLE is estimated from Vals MMLU-Pro with a inverse Michaelis–Menten curve fitted on 29 models measured on both: y = 0.08531·(x − 0.0000) / (0.0000 + 1.0439 − x), R² = 0.48, cross-validated error 9.4 pp. It is used for 15 estimates.
| Estimated model | Vals MMLU-Pro | HLE | Source |
|---|---|---|---|
| Gemini 3 Flash | 88.6% | 47.9% | estimated ± 9.4 pp, low confidence |
| Muse Spark 1.2 | 88.3% | 46.8% | estimated ± 9.4 pp, low confidence |
| GLM-5.3 | 86.8% | 42.1% | estimated ± 9.4 pp, low confidence |
| Gemini 3.1 Flash-Lite | 86.2% | 40.4% | estimated ± 9.4 pp, low confidence |
| GLM-5.3-Flash | 86.1% | 40.1% | estimated ± 9.4 pp, low confidence |
| Gemini 3.5 Flash-Lite | 85.8% | 39.4% | estimated ± 9.4 pp, low confidence |
| MiniMax M3 | 84.2% | 35.6% | estimated ± 9.4 pp, low confidence |
| Qwen3.5 Flash | 84.1% | 35.4% | estimated ± 9.4 pp, low confidence |
| MiMo-V2.5 | 82.9% | 32.9% | estimated ± 9.4 pp, low confidence |
| GLM-4.6 | 82.2% | 31.6% | estimated ± 9.4 pp, low confidence |
| GLM-4.5 | 81.2% | 29.9% | estimated ± 9.4 pp, low confidence |
| Claude Haiku 4.5 | 78.7% | 26.1% | estimated ± 9.4 pp, low confidence |
| Mistral Medium 3.5 128B | 75.3% | 22.1% | estimated ± 9.4 pp, low confidence |
| Laguna XS.2 | 69.1% | 16.7% | estimated ± 9.4 pp, low confidence |
| Laguna M.1 | 68.8% | 16.5% | estimated ± 9.4 pp, low confidence |
Fitted on
| Model | Vals MMLU-Pro | HLE |
|---|---|---|
| Claude Fable 5.1 | 92.4% | 65.0% |
| Claude Opus 4.8 | 89.6% | 57.9% |
| Claude Opus 5 | 91.6% | 64.7% |
| Claude Sonnet 4.6 | 87.3% | 49.0% |
| Claude Sonnet 5 | 87.5% | 57.4% |
| DeepSeek V4 Flash 0731 | 86.2% | 34.8% |
| DeepSeek V4 Pro 0813 | 87.0% | 42.7% |
| Gemini 3.5 Flash | 89.5% | 40.2% |
| GLM-4.7 | 82.7% | 24.8% |
| GLM-5.1 | 86.9% | 31.0% |
| GLM-5.2 | 86.7% | 40.5% |
| GPT-5.4 mini | 84.6% | 41.5% |
| GPT-5.4 nano | 77.2% | 37.7% |
| GPT-5.5 | 88.1% | 52.2% |
| Grok 4.3 | 85.8% | 35.0% |
| Inkling | 86.3% | 46.0% |
| Inkling-Small | 85.6% | 47.8% |
| Kimi K2.6 | 87.6% | 34.7% |
| Kimi K3 | 88.0% | 56.0% |
| Ling 3.0 Flash | 82.0% | 22.7% |
| MiMo-V2.5-Pro | 84.6% | 48.0% |
| MiniMax M2.7 | 80.4% | 28.0% |
| Muse Spark | 87.3% | 50.4% |
| Muse Spark 1.1 | 88.7% | 62.1% |
| Nemotron 3 Ultra | 85.8% | 26.7% |
| Qwen3.6 Plus | 87.7% | 28.8% |
| Qwen3.7 Max | 89.3% | 41.4% |
| Qwen3.8-27B | 84.3% | 30.8% |
| Qwen3.8 Max | 88.6% | 43.6% |