Calibration
AA-Omniscience Accuracy → Vals MMLU-Pro
Vals MMLU-Pro is estimated from AA-Omniscience Accuracy with a linear curve fitted on 46 models measured on both: y = 0.1855·x + 0.7941, R² = 0.55, cross-validated error 2.2 pp. It is used for 79 estimates.
| Estimated model | AA-Omniscience Accuracy | Vals MMLU-Pro | Source |
|---|---|---|---|
| Apodex 1.1 | 31.7% | 85.3% | estimated ± 2.2 pp, high confidence |
| Apodex 1.1 Mini | 31.7% | 85.3% | estimated ± 2.2 pp, high confidence |
| Celeris-1 | 11.0% | 81.5% | estimated ± 2.2 pp, medium confidence |
| Claude 3 Haiku | 17.6% | 82.7% | estimated ± 2.2 pp, high confidence |
| Claude 4 Sonnet | 22.7% | 83.6% | estimated ± 2.2 pp, high confidence |
| Command A+ | 8.9% | 81.1% | estimated ± 2.2 pp, medium confidence |
| DeepSeek-R1 | 30.5% | 85.1% | estimated ± 2.2 pp, high confidence |
| DeepSeek V3 0324 | 24.3% | 83.9% | estimated ± 2.2 pp, high confidence |
| DeepSeek V3.1 | 23.1% | 83.7% | estimated ± 2.2 pp, high confidence |
| DeepSeek V3.1 (Reasoning) | 29.0% | 84.8% | estimated ± 2.2 pp, high confidence |
| DeepSeek V3.2 | 24.0% | 83.9% | estimated ± 2.2 pp, high confidence |
| Exaone 4.0 1.2B | 5.0% | 80.3% | estimated ± 2.2 pp, medium confidence |
| Exaone 4.0 32B | 10.6% | 81.4% | estimated ± 2.2 pp, medium confidence |
| Gemini 2.5 Flash | 26.1% | 84.3% | estimated ± 2.2 pp, high confidence |
| Gemma 3 27B | 13.0% | 81.8% | estimated ± 2.2 pp, medium confidence |
| GLM-4.5-Air | 16.3% | 82.4% | estimated ± 2.2 pp, high confidence |
| GLM-5-Turbo | 28.4% | 84.7% | estimated ± 2.2 pp, high confidence |
| GLM-5V-Turbo | 29.3% | 84.8% | estimated ± 2.2 pp, high confidence |
| GPT-4o | 19.9% | 83.1% | estimated ± 2.2 pp, high confidence |
| GPT-5.1-Codex | 39.9% | 86.8% | estimated ± 2.2 pp, high confidence |
| GPT-5.1-Codex-Max | 39.9% | 86.8% | estimated ± 2.2 pp, high confidence |
| GPT-5.2-Codex | 41.1% | 87.0% | estimated ± 2.2 pp, high confidence |
| GPT-5.3 Codex | 52.9% | 89.2% | estimated ± 2.2 pp, high confidence |
| GPT-5 (high) | 40.3% | 86.9% | estimated ± 2.2 pp, high confidence |
| GPT-5 (medium) | 39.5% | 86.7% | estimated ± 2.2 pp, high confidence |
| GPT-OSS 120B | 21.8% | 83.5% | estimated ± 2.2 pp, high confidence |
| GPT-OSS 20B | 16.0% | 82.4% | estimated ± 2.2 pp, high confidence |
| Granite-4.0-350M | 3.9% | 80.1% | estimated ± 2.2 pp, medium confidence |
| Granite-4.0-H-1B | 5.2% | 80.4% | estimated ± 2.2 pp, medium confidence |
| Granite-4.0-H-350M | 3.8% | 80.1% | estimated ± 2.2 pp, medium confidence |
| Grok 4 | 40.5% | 86.9% | estimated ± 2.2 pp, high confidence |
| Grok 4.1 Fast | 17.2% | 82.6% | estimated ± 2.2 pp, high confidence |
| Grok 4.1 Fast (Reasoning) | 25.1% | 84.1% | estimated ± 2.2 pp, high confidence |
| Grok 4 Fast (Reasoning) | 22.8% | 83.6% | estimated ± 2.2 pp, high confidence |
| Grok Code Fast 1 | 23.5% | 83.8% | estimated ± 2.2 pp, high confidence |
| Hy3 | 32.0% | 85.3% | estimated ± 2.2 pp, high confidence |
| K-Exaone | 16.4% | 82.5% | estimated ± 2.2 pp, high confidence |
| Kimi K2 | 27.4% | 84.5% | estimated ± 2.2 pp, high confidence |
| Kimi K2.7 Code | 39.6% | 86.8% | estimated ± 2.2 pp, high confidence |
| LFM2.5-2.6B | 4.4% | 80.2% | estimated ± 2.2 pp, medium confidence |
| LFM2.5-8B-A1B | 9.4% | 81.2% | estimated ± 2.2 pp, medium confidence |
| LFM2.5-VL-1.6B-Extract | 5.8% | 80.5% | estimated ± 2.2 pp, medium confidence |
| Ling 3.0 Flash VL | 14.4% | 82.1% | estimated ± 2.2 pp, medium confidence |
| Ling 3.0 Tiny | 8.5% | 81.0% | estimated ± 2.2 pp, medium confidence |
| Llama 3.1 405B | 23.2% | 83.7% | estimated ± 2.2 pp, high confidence |
| Llama 4 Maverick | 24.9% | 84.0% | estimated ± 2.2 pp, high confidence |
| Llama 4 Scout | 15.2% | 82.2% | estimated ± 2.2 pp, medium confidence |
| MiMo-V2.6-Flash | 27.0% | 84.4% | estimated ± 2.2 pp, high confidence |
| MiMo-V2.6-Pro | 34.8% | 85.9% | estimated ± 2.2 pp, high confidence |
| MiMo-V2-Omni | 19.3% | 83.0% | estimated ± 2.2 pp, high confidence |
| MiMo-V2-Pro | 26.6% | 84.3% | estimated ± 2.2 pp, high confidence |
| Mistral Large 2 | 19.9% | 83.1% | estimated ± 2.2 pp, high confidence |
| Mistral Large 3 | 25.0% | 84.1% | estimated ± 2.2 pp, high confidence |
| Mistral Large 4 | 25.8% | 84.2% | estimated ± 2.2 pp, high confidence |
| Mistral Medium 3 | 18.3% | 82.8% | estimated ± 2.2 pp, high confidence |
| Mistral Small 4 | 21.7% | 83.4% | estimated ± 2.2 pp, high confidence |
| Mistral Small 4 (Reasoning) | 21.7% | 83.4% | estimated ± 2.2 pp, high confidence |
| Muse Glimmer 30B | 27.0% | 84.4% | estimated ± 2.2 pp, high confidence |
| Muse Spark 1.3 | 43.6% | 87.5% | estimated ± 2.2 pp, high confidence |
| Nemotron 3 Nano 30B | 17.3% | 82.6% | estimated ± 2.2 pp, high confidence |
| Nemotron 3 Super 100B | 24.3% | 83.9% | estimated ± 2.2 pp, high confidence |
| Nemotron Ultra 253B | 20.1% | 83.1% | estimated ± 2.2 pp, high confidence |
| North Mini Code | 18.9% | 82.9% | estimated ± 2.2 pp, high confidence |
| Nova Pro | 16.9% | 82.5% | estimated ± 2.2 pp, high confidence |
| o3 | 38.6% | 86.6% | estimated ± 2.2 pp, high confidence |
| Phi-4 | 14.1% | 82.0% | estimated ± 2.2 pp, medium confidence |
| Quasar 438B | 15.5% | 82.3% | estimated ± 2.2 pp, medium confidence |
| Qwen3.5 397B (Reasoning) | 24.5% | 84.0% | estimated ± 2.2 pp, high confidence |
| Qwen 3.6 Max (preview) | 37.9% | 86.4% | estimated ± 2.2 pp, high confidence |
| Qwen3.8 Max Preview | 31.7% | 85.3% | estimated ± 2.2 pp, high confidence |
| Qwen3 Max | 24.4% | 83.9% | estimated ± 2.2 pp, high confidence |
| Qwen3-Omni-30B-A3B-Instruct | 14.3% | 82.1% | estimated ± 2.2 pp, medium confidence |
| Qwen3-Omni-30B-A3B-Thinking | 14.6% | 82.1% | estimated ± 2.2 pp, medium confidence |
| Sarvam 105B | 17.6% | 82.7% | estimated ± 2.2 pp, high confidence |
| Sarvam 30B | 12.6% | 81.8% | estimated ± 2.2 pp, medium confidence |
| Solar Pro 2 | 16.1% | 82.4% | estimated ± 2.2 pp, high confidence |
| Solar Pro 3 | 18.5% | 82.8% | estimated ± 2.2 pp, high confidence |
| Step 3.7 Flash | 25.8% | 84.2% | estimated ± 2.2 pp, high confidence |
| Ultravox v0.6 Llama 3.3 70B | 19.0% | 82.9% | estimated ± 2.2 pp, high confidence |