Calibration
AA-Omniscience Accuracy → MMLU
MMLU is estimated from AA-Omniscience Accuracy with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.0183·x / (0.03628 + x), R² = 0.98, cross-validated error 1.1 pp. It is used for 178 estimates.
| Estimated model | AA-Omniscience Accuracy | MMLU | Source |
|---|---|---|---|
| A.X K2 | 18.6% | 85.2% | estimated ± 1.1 pp, medium confidence |
| Apodex 1.1 | 31.7% | 91.4% | estimated ± 1.1 pp, medium confidence |
| Apodex 1.1 Mini | 31.7% | 91.4% | estimated ± 1.1 pp, medium confidence |
| Celeris-1 | 11.0% | 76.6% | estimated ± 1.1 pp, low confidence |
| Claude 3 Haiku | 17.6% | 84.4% | estimated ± 1.1 pp, medium confidence |
| Claude 4 Sonnet | 22.7% | 87.8% | estimated ± 1.1 pp, medium confidence |
| Claude Fable 5 | 65.4% | 96.5% | estimated ± 1.1 pp, low confidence |
| Claude Fable 5.1 | 67.2% | 96.6% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.5 | 40.9% | 93.5% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.5 Thinking | 46.6% | 94.5% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.6 | 45.8% | 94.4% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 47.0% | 94.5% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.7 | 44.7% | 94.2% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 48.9% | 94.8% | estimated ± 1.1 pp, low confidence |
| Claude Opus 4.8 | 48.8% | 94.8% | estimated ± 1.1 pp, low confidence |
| Claude Opus 5 | 60.9% | 96.1% | estimated ± 1.1 pp, low confidence |
| Claude Opus 5.5 | 66.2% | 96.5% | estimated ± 1.1 pp, low confidence |
| Claude Sonnet 4.6 | 38.6% | 93.1% | estimated ± 1.1 pp, low confidence |
| Claude Sonnet 5 | 40.1% | 93.4% | estimated ± 1.1 pp, low confidence |
| Claude Sonnet 5.5 | 54.0% | 95.4% | estimated ± 1.1 pp, low confidence |
| Command A+ | 8.9% | 72.3% | estimated ± 1.1 pp, low confidence |
| DeepSeek-R1 | 30.5% | 91.0% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V3 | 25.5% | 89.1% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V3 0324 | 24.3% | 88.6% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V3.1 | 23.1% | 88.0% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V3.1 (Reasoning) | 29.0% | 90.5% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V3.2 | 24.0% | 88.5% | estimated ± 1.1 pp, medium confidence |
| DeepSeek V4.1 Flash | 46.4% | 94.4% | estimated ± 1.1 pp, low confidence |
| DeepSeek V4 Flash 0731 | 40.4% | 93.4% | estimated ± 1.1 pp, low confidence |
| DeepSeek V4 Pro 0813 | 49.1% | 94.8% | estimated ± 1.1 pp, low confidence |
| Exaone 4.0 1.2B | 5.0% | 59.0% | estimated ± 1.1 pp, low confidence |
| Exaone 4.0 32B | 10.6% | 75.9% | estimated ± 1.1 pp, low confidence |
| Gemini 2.5 Flash | 26.1% | 89.4% | estimated ± 1.1 pp, medium confidence |
| Gemini 2.5 Pro | 39.1% | 93.2% | estimated ± 1.1 pp, low confidence |
| Gemini 3.1 Pro | 54.9% | 95.5% | estimated ± 1.1 pp, low confidence |
| Gemini 3.5 Flash | 51.9% | 95.2% | estimated ± 1.1 pp, low confidence |
| Gemini 3.5 Flash-Lite | 29.5% | 90.7% | estimated ± 1.1 pp, medium confidence |
| Gemini 3.6 Flash | 50.0% | 94.9% | estimated ± 1.1 pp, low confidence |
| Gemini 3.7 Flash | 55.3% | 95.6% | estimated ± 1.1 pp, low confidence |
| Gemini 3.8 Flash | 54.6% | 95.5% | estimated ± 1.1 pp, low confidence |
| Gemini 3 Flash | 45.8% | 94.4% | estimated ± 1.1 pp, low confidence |
| Gemini 3 Pro | 55.8% | 95.6% | estimated ± 1.1 pp, low confidence |
| Gemini 4 Argon | 49.9% | 94.9% | estimated ± 1.1 pp, low confidence |
| Gemma 3 27B | 13.0% | 79.6% | estimated ± 1.1 pp, low confidence |
| Gemma 4 12B | 15.6% | 82.6% | estimated ± 1.1 pp, medium confidence |
| Gemma 4 26B A4B | 19.1% | 85.6% | estimated ± 1.1 pp, medium confidence |
| Gemma 4 31B | 20.0% | 86.2% | estimated ± 1.1 pp, medium confidence |
| Gemma 4 E2B | 6.6% | 65.7% | estimated ± 1.1 pp, low confidence |
| Gemma 4 E4B | 8.6% | 71.6% | estimated ± 1.1 pp, low confidence |
| GLM-4.5-Air | 16.3% | 83.3% | estimated ± 1.1 pp, medium confidence |
| GLM-4.6 | 21.4% | 87.1% | estimated ± 1.1 pp, medium confidence |
| GLM-4.7 | 29.3% | 90.6% | estimated ± 1.1 pp, medium confidence |
| GLM-5 | 26.3% | 89.5% | estimated ± 1.1 pp, medium confidence |
| GLM-5.1 | 23.7% | 88.3% | estimated ± 1.1 pp, medium confidence |
| GLM-5.2 | 24.3% | 88.6% | estimated ± 1.1 pp, medium confidence |
| GLM-5.3 | 33.9% | 92.0% | estimated ± 1.1 pp, medium confidence |
| GLM-5-Turbo | 28.4% | 90.3% | estimated ± 1.1 pp, medium confidence |
| GLM-5V-Turbo | 29.3% | 90.6% | estimated ± 1.1 pp, medium confidence |
| GPT-4o | 19.9% | 86.1% | estimated ± 1.1 pp, medium confidence |
| GPT-5.1 | 37.7% | 92.9% | estimated ± 1.1 pp, low confidence |
| GPT-5.1-Codex | 39.9% | 93.3% | estimated ± 1.1 pp, low confidence |
| GPT-5.1-Codex-Max | 39.9% | 93.3% | estimated ± 1.1 pp, low confidence |
| GPT-5.2 | 44.3% | 94.1% | estimated ± 1.1 pp, low confidence |
| GPT-5.2-Codex | 41.1% | 93.6% | estimated ± 1.1 pp, low confidence |
| GPT-5.3 Codex | 52.9% | 95.3% | estimated ± 1.1 pp, low confidence |
| GPT-5.4 | 50.8% | 95.0% | estimated ± 1.1 pp, low confidence |
| GPT-5.4 mini | 37.5% | 92.8% | estimated ± 1.1 pp, low confidence |
| GPT-5.4 nano | 25.7% | 89.2% | estimated ± 1.1 pp, medium confidence |
| GPT-5.5 | 58.0% | 95.8% | estimated ± 1.1 pp, low confidence |
| GPT-5.6 Luna | 42.7% | 93.9% | estimated ± 1.1 pp, low confidence |
| GPT-5.6 Sol | 59.4% | 96.0% | estimated ± 1.1 pp, low confidence |
| GPT-5.6 Terra | 46.8% | 94.5% | estimated ± 1.1 pp, low confidence |
| GPT-5 (high) | 40.3% | 93.4% | estimated ± 1.1 pp, low confidence |
| GPT-5 (medium) | 39.5% | 93.3% | estimated ± 1.1 pp, low confidence |
| GPT-6.1 Sol | 62.1% | 96.2% | estimated ± 1.1 pp, low confidence |
| GPT-6 Astra | 62.6% | 96.3% | estimated ± 1.1 pp, low confidence |
| GPT-6 Luna | 43.8% | 94.0% | estimated ± 1.1 pp, low confidence |
| GPT-6 Sol | 54.5% | 95.5% | estimated ± 1.1 pp, low confidence |
| GPT-OSS 120B | 21.8% | 87.3% | estimated ± 1.1 pp, medium confidence |
| GPT-OSS 20B | 16.0% | 83.0% | estimated ± 1.1 pp, medium confidence |
| Granite-4.0-350M | 3.9% | 52.8% | estimated ± 1.1 pp, low confidence |
| Granite-4.0-H-1B | 5.2% | 60.0% | estimated ± 1.1 pp, low confidence |
| Granite-4.0-H-350M | 3.8% | 52.1% | estimated ± 1.1 pp, low confidence |
| Granite 4.2 30B | 10.1% | 74.9% | estimated ± 1.1 pp, low confidence |
| Granite 4.2 3B | 9.2% | 73.0% | estimated ± 1.1 pp, low confidence |
| Granite 4.2 8B | 11.2% | 76.9% | estimated ± 1.1 pp, low confidence |
| Grok 4 | 40.5% | 93.5% | estimated ± 1.1 pp, low confidence |
| Grok 4.1 Fast | 17.2% | 84.1% | estimated ± 1.1 pp, medium confidence |
| Grok 4.1 Fast (Reasoning) | 25.1% | 89.0% | estimated ± 1.1 pp, medium confidence |
| Grok 4.3 | 34.6% | 92.2% | estimated ± 1.1 pp, low confidence |
| Grok 4.5 | 51.6% | 95.1% | estimated ± 1.1 pp, low confidence |
| Grok 4.6 | 48.2% | 94.7% | estimated ± 1.1 pp, low confidence |
| Grok 4.7 | 47.4% | 94.6% | estimated ± 1.1 pp, low confidence |
| Grok 4 Fast (Reasoning) | 22.8% | 87.9% | estimated ± 1.1 pp, medium confidence |
| Grok Code Fast 1 | 23.5% | 88.2% | estimated ± 1.1 pp, medium confidence |
| Hy3 | 32.0% | 91.5% | estimated ± 1.1 pp, medium confidence |
| Hy3 Preview | 31.5% | 91.3% | estimated ± 1.1 pp, medium confidence |
| Inkling | 41.6% | 93.7% | estimated ± 1.1 pp, low confidence |
| Inkling-Small | 33.2% | 91.8% | estimated ± 1.1 pp, medium confidence |
| K-Exaone | 16.4% | 83.4% | estimated ± 1.1 pp, medium confidence |
| K-EXAONE 2.0 | 13.1% | 79.7% | estimated ± 1.1 pp, low confidence |
| Kimi K2.6 | 32.6% | 91.6% | estimated ± 1.1 pp, medium confidence |
| Kimi K2 | 27.4% | 89.9% | estimated ± 1.1 pp, medium confidence |
| Kimi K2.5 | 35.2% | 92.3% | estimated ± 1.1 pp, low confidence |
| Kimi K2.5 (Reasoning) | 35.2% | 92.3% | estimated ± 1.1 pp, low confidence |
| Kimi K2.7 Code | 39.6% | 93.3% | estimated ± 1.1 pp, low confidence |
| Kimi K3 | 47.6% | 94.6% | estimated ± 1.1 pp, low confidence |
| LFM2.5-2.6B | 4.4% | 55.8% | estimated ± 1.1 pp, low confidence |
| LFM2.5-8B-A1B | 9.4% | 73.5% | estimated ± 1.1 pp, low confidence |
| LFM2.5-VL-1.6B-Extract | 5.8% | 62.6% | estimated ± 1.1 pp, low confidence |
| Ling 2.6 Flash | 15.6% | 82.6% | estimated ± 1.1 pp, medium confidence |
| Ling 3.0 Flash | 18.2% | 84.9% | estimated ± 1.1 pp, medium confidence |
| Ling 3.0 Flash FP8 | 18.2% | 84.9% | estimated ± 1.1 pp, medium confidence |
| Ling 3.0 Flash VL | 14.4% | 81.3% | estimated ± 1.1 pp, medium confidence |
| Ling 3.0 Tiny | 8.5% | 71.4% | estimated ± 1.1 pp, low confidence |
| Ling 3.1 Flash | 29.1% | 90.5% | estimated ± 1.1 pp, medium confidence |
| Llama 3.1 405B | 23.2% | 88.1% | estimated ± 1.1 pp, medium confidence |
| Llama 4 Maverick | 24.9% | 88.9% | estimated ± 1.1 pp, medium confidence |
| Llama 4 Scout | 15.2% | 82.2% | estimated ± 1.1 pp, medium confidence |
| Mercury 2.5 | 22.0% | 87.4% | estimated ± 1.1 pp, medium confidence |
| MiMo-V2.5-Pro | 22.4% | 87.6% | estimated ± 1.1 pp, medium confidence |
| MiMo-V2.6-Flash | 27.0% | 89.8% | estimated ± 1.1 pp, medium confidence |
| MiMo-V2.6-Pro | 34.8% | 92.2% | estimated ± 1.1 pp, low confidence |
| MiMo-V2-Flash | 15.6% | 82.6% | estimated ± 1.1 pp, medium confidence |
| MiMo-V2-Omni | 19.3% | 85.7% | estimated ± 1.1 pp, medium confidence |
| MiMo-V2-Pro | 26.6% | 89.6% | estimated ± 1.1 pp, medium confidence |
| MiniCPM5-2B | 8.4% | 71.1% | estimated ± 1.1 pp, low confidence |
| MiniMax M2.7 | 26.8% | 89.7% | estimated ± 1.1 pp, medium confidence |
| MiniMax M3 | 16.7% | 83.7% | estimated ± 1.1 pp, medium confidence |
| Mistral Large 2 | 19.9% | 86.1% | estimated ± 1.1 pp, medium confidence |
| Mistral Large 3 | 25.0% | 88.9% | estimated ± 1.1 pp, medium confidence |
| Mistral Large 4 | 25.8% | 89.3% | estimated ± 1.1 pp, medium confidence |
| Mistral Medium 3 | 18.3% | 85.0% | estimated ± 1.1 pp, medium confidence |
| Mistral Medium 3.5 128B | 24.7% | 88.8% | estimated ± 1.1 pp, medium confidence |
| Mistral Small 4 | 21.7% | 87.2% | estimated ± 1.1 pp, medium confidence |
| Mistral Small 4 (Reasoning) | 21.7% | 87.2% | estimated ± 1.1 pp, medium confidence |
| Muse Glimmer 30B | 27.0% | 89.8% | estimated ± 1.1 pp, medium confidence |
| Muse Spark | 49.6% | 94.9% | estimated ± 1.1 pp, low confidence |
| Muse Spark 1.1 | 52.1% | 95.2% | estimated ± 1.1 pp, low confidence |
| Muse Spark 1.2 | 45.4% | 94.3% | estimated ± 1.1 pp, low confidence |
| Muse Spark 1.3 | 43.6% | 94.0% | estimated ± 1.1 pp, low confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 14.4% | 81.3% | estimated ± 1.1 pp, medium confidence |
| Nemotron 3 Nano 30B | 17.3% | 84.2% | estimated ± 1.1 pp, medium confidence |
| Nemotron 3 Nano Omni 30B A3B | 15.2% | 82.2% | estimated ± 1.1 pp, medium confidence |
| Nemotron 3 Super 100B | 24.3% | 88.6% | estimated ± 1.1 pp, medium confidence |
| Nemotron 3 Ultra | 21.6% | 87.2% | estimated ± 1.1 pp, medium confidence |
| Nemotron Ultra 253B | 20.1% | 86.3% | estimated ± 1.1 pp, medium confidence |
| North Mini Code | 18.9% | 85.4% | estimated ± 1.1 pp, medium confidence |
| Nova Pro | 16.9% | 83.8% | estimated ± 1.1 pp, medium confidence |
| o3 | 38.6% | 93.1% | estimated ± 1.1 pp, low confidence |
| Phi-4 | 14.1% | 81.0% | estimated ± 1.1 pp, medium confidence |
| Quasar 438B | 15.5% | 82.5% | estimated ± 1.1 pp, medium confidence |
| Qwen3.5-122B-A10B | 24.4% | 88.6% | estimated ± 1.1 pp, medium confidence |
| Qwen3.5-27B | 20.7% | 86.6% | estimated ± 1.1 pp, medium confidence |
| Qwen3.5-35B-A3B | 20.1% | 86.3% | estimated ± 1.1 pp, medium confidence |
| Qwen3.5 397B | 24.5% | 88.7% | estimated ± 1.1 pp, medium confidence |
| Qwen3.5 397B (Reasoning) | 24.5% | 88.7% | estimated ± 1.1 pp, medium confidence |
| Qwen3.6-27B | 19.6% | 85.9% | estimated ± 1.1 pp, medium confidence |
| Qwen3.6-35B-A3B | 18.8% | 85.4% | estimated ± 1.1 pp, medium confidence |
| Qwen 3.6 Max (preview) | 37.9% | 92.9% | estimated ± 1.1 pp, low confidence |
| Qwen3.6 Plus | 26.4% | 89.5% | estimated ± 1.1 pp, medium confidence |
| Qwen3.7 Max | 31.1% | 91.2% | estimated ± 1.1 pp, medium confidence |
| Qwen3.7 Plus | 22.5% | 87.7% | estimated ± 1.1 pp, medium confidence |
| Qwen3.8-27B | 15.6% | 82.6% | estimated ± 1.1 pp, medium confidence |
| Qwen3.8-Flash-Next | 24.5% | 88.7% | estimated ± 1.1 pp, medium confidence |
| Qwen3.8 Max Preview | 31.7% | 91.4% | estimated ± 1.1 pp, medium confidence |
| Qwen3 Max | 24.4% | 88.6% | estimated ± 1.1 pp, medium confidence |
| Qwen3-Omni-30B-A3B-Instruct | 14.3% | 81.2% | estimated ± 1.1 pp, medium confidence |
| Qwen3-Omni-30B-A3B-Thinking | 14.6% | 81.6% | estimated ± 1.1 pp, medium confidence |
| Sarvam 105B | 17.6% | 84.4% | estimated ± 1.1 pp, medium confidence |
| Sarvam 30B | 12.6% | 79.1% | estimated ± 1.1 pp, low confidence |
| Solar Pro 2 | 16.1% | 83.1% | estimated ± 1.1 pp, medium confidence |
| Solar Pro 3 | 18.5% | 85.1% | estimated ± 1.1 pp, medium confidence |
| Solar Pro 4 | 18.9% | 85.4% | estimated ± 1.1 pp, medium confidence |
| Step 3.7 Flash | 25.8% | 89.3% | estimated ± 1.1 pp, medium confidence |
| Step 5 Preview | 41.5% | 93.6% | estimated ± 1.1 pp, low confidence |
| Trinity-Large-Thinking | 22.5% | 87.7% | estimated ± 1.1 pp, medium confidence |
| Ultravox v0.6 Llama 3.3 70B | 19.0% | 85.5% | estimated ± 1.1 pp, medium confidence |