Calibration
AA-Omniscience Accuracy → MMLU
MMLU is estimated from AA-Omniscience Accuracy with a offset logistic curve fitted on 10 models measured on both: y = 0.8033 + (0.9047 − 0.8033) / (1 + exp(−146.94·(x − 0.1970))), R² = 0.98, cross-validated error 0.9 pp. It is used for 144 estimates.
| Estimated model | AA-Omniscience Accuracy | MMLU | Source |
|---|---|---|---|
| Claude Opus 4.5 Thinking | 46.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.6 | 45.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.6 (Adaptive) | 47.0% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.7 | 48.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.8 | 48.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Sonnet 5.5 | 51.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| DeepSeek V4.1 Flash 1017 | 46.4% | 90.5% | estimated ± 0.9 pp, medium confidence |
| DeepSeek V4 Pro | 49.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 3.5 Flash | 51.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 3.6 Flash | 50.0% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 3.7 Flash | 55.3% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 3.8 Flash | 50.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 3 Flash | 45.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 4 Argon | 48.3% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 4 Hydrogen | 50.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.2 | 44.3% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.4 | 50.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.6 Terra | 46.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-6.1 Terra | 47.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-6 Sol | 54.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Grok 4.5 | 51.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Grok 4.6 | 48.2% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Kimi K3.1 Pro | 48.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Kimi K3 | 47.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| MiniMax M3 | 44.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| MiniMax M3.1 Flash | 45.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Spark | 49.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Spark 1.1 | 52.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Spark 1.2 | 45.4% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Qwen4 Max | 43.7% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Spark 1.3 | 43.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.6 Luna | 42.7% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Grok 4.7 | 42.7% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Spark Open | 42.0% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.2-Codex | 41.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Wonder Full | 41.0% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.5 | 40.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Grok 4 | 40.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| DeepSeek V4 Flash 0731 | 40.4% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5 | 40.3% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Sonnet 5 | 40.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.1-Codex | 39.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.1-Codex-Max | 39.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-6 Luna | 39.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Inkling | 39.3% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Gemini 2.5 Pro | 39.1% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Sonnet 4.6 | 38.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Muse Wonder | 38.0% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Ling 3.0 Tiny | 37.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Qwen 3.6 Max (preview) | 37.9% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.1 | 37.7% | 90.5% | estimated ± 0.9 pp, medium confidence |
| GPT-5.4 mini | 37.5% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Step 5 Preview | 36.4% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Kimi K2.5 | 35.2% | 90.5% | estimated ± 0.9 pp, medium confidence |
| MiMo-V2.6-Pro | 34.8% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Claude Haiku 5.5 | 34.7% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Grok 4.3 | 34.6% | 90.5% | estimated ± 0.9 pp, medium confidence |
| Inkling 2 Small | 33.8% | 90.5% | estimated ± 0.9 pp, high confidence |
| Kimi K2.6 | 32.6% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-5.3 | 32.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| Inkling-Small | 31.8% | 90.5% | estimated ± 0.9 pp, high confidence |
| Hy3 Preview | 31.5% | 90.5% | estimated ± 0.9 pp, high confidence |
| Qwen3.7 Max | 31.1% | 90.5% | estimated ± 0.9 pp, high confidence |
| Beam 1.0 501B-A23B | 30.7% | 90.5% | estimated ± 0.9 pp, high confidence |
| Qwen3.8 Max Preview | 29.5% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-4.7 | 29.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-5V-Turbo | 29.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-5-Turbo | 28.4% | 90.5% | estimated ± 0.9 pp, high confidence |
| Apodex 1.1 | 28.2% | 90.5% | estimated ± 0.9 pp, high confidence |
| Apodex 1.1 Mini | 28.2% | 90.5% | estimated ± 0.9 pp, high confidence |
| Claude 3.7 Sonnet | 28.2% | 90.5% | estimated ± 0.9 pp, high confidence |
| Claude 3.7 Sonnet Thinking | 28.0% | 90.5% | estimated ± 0.9 pp, high confidence |
| Muse Glimmer 30B | 28.0% | 90.5% | estimated ± 0.9 pp, high confidence |
| MiMo-V2.6-Flash | 27.2% | 90.5% | estimated ± 0.9 pp, high confidence |
| MiniMax M2.7 | 26.8% | 90.5% | estimated ± 0.9 pp, high confidence |
| MiMo-V2-Pro | 26.6% | 90.5% | estimated ± 0.9 pp, high confidence |
| Gemini 3.5 Flash-Lite | 26.5% | 90.5% | estimated ± 0.9 pp, high confidence |
| Qwen3.6 Plus | 26.4% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-5 | 26.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| Nemotron 3.5 Super VL 120B A12B | 26.2% | 90.5% | estimated ± 0.9 pp, high confidence |
| Gemini 2.5 Flash | 26.1% | 90.5% | estimated ± 0.9 pp, high confidence |
| Mistral Large 4 | 26.0% | 90.5% | estimated ± 0.9 pp, high confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 25.9% | 90.5% | estimated ± 0.9 pp, high confidence |
| Step 3.7 Flash | 25.8% | 90.5% | estimated ± 0.9 pp, high confidence |
| GPT-5.4 nano | 25.7% | 90.5% | estimated ± 0.9 pp, high confidence |
| DeepSeek V3 | 25.5% | 90.5% | estimated ± 0.9 pp, high confidence |
| Grok 4.1 Fast (Reasoning) | 25.1% | 90.5% | estimated ± 0.9 pp, high confidence |
| Mistral Large 3 | 25.0% | 90.5% | estimated ± 0.9 pp, high confidence |
| Qwen3 Max | 24.4% | 90.5% | estimated ± 0.9 pp, high confidence |
| DeepSeek V3 0324 | 24.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| GLM-5.2 | 24.3% | 90.5% | estimated ± 0.9 pp, high confidence |
| DeepSeek V3.2 | 24.0% | 90.4% | estimated ± 0.9 pp, high confidence |
| GLM-5.1 | 23.7% | 90.4% | estimated ± 0.9 pp, high confidence |
| Gemini 2.0 Flash | 23.5% | 90.4% | estimated ± 0.9 pp, high confidence |
| Grok Code Fast 1 | 23.5% | 90.4% | estimated ± 0.9 pp, high confidence |
| Qwen3.8-Flash-Next | 23.5% | 90.4% | estimated ± 0.9 pp, high confidence |
| DeepSeek V3.1 | 23.1% | 90.4% | estimated ± 0.9 pp, high confidence |
| Mistral Medium 3.5 128B | 23.0% | 90.4% | estimated ± 0.9 pp, high confidence |
| Grok 4 Fast (Reasoning) | 22.8% | 90.4% | estimated ± 0.9 pp, high confidence |
| Claude 4 Sonnet | 22.7% | 90.3% | estimated ± 0.9 pp, high confidence |
| MiMo-V2.5-Pro | 22.4% | 90.3% | estimated ± 0.9 pp, high confidence |
| Mercury 2.5 | 22.0% | 90.1% | estimated ± 0.9 pp, high confidence |
| Nemotron 3.5 Bolt 74B-A9B BF16 | 21.8% | 90.0% | estimated ± 0.9 pp, high confidence |
| Nemotron 3 Ultra | 21.6% | 89.9% | estimated ± 0.9 pp, high confidence |
| GLM-4.6 | 21.4% | 89.7% | estimated ± 0.9 pp, high confidence |
| Mistral Small 4 | 21.4% | 89.7% | estimated ± 0.9 pp, high confidence |
| Muse Ember | 21.1% | 89.3% | estimated ± 0.9 pp, high confidence |
| Qwen3.5-27B | 20.7% | 88.6% | estimated ± 0.9 pp, high confidence |
| Magistral Medium | 20.3% | 87.5% | estimated ± 0.9 pp, high confidence |
| Qwen3.5-35B-A3B | 20.1% | 86.9% | estimated ± 0.9 pp, high confidence |
| Gemma 4 31B | 19.9% | 86.2% | estimated ± 0.9 pp, high confidence |
| GPT-4o | 19.9% | 86.2% | estimated ± 0.9 pp, high confidence |
| Mistral Large 2 | 19.9% | 86.2% | estimated ± 0.9 pp, high confidence |
| Qwen3.6-27B | 19.6% | 85.0% | estimated ± 0.9 pp, high confidence |
| MiMo-V2-Omni | 19.3% | 84.0% | estimated ± 0.9 pp, high confidence |
| Solar Pro 3 | 18.5% | 81.8% | estimated ± 0.9 pp, high confidence |
| North Mini Code | 18.4% | 81.6% | estimated ± 0.9 pp, high confidence |
| Solar Pro 4 | 18.4% | 81.6% | estimated ± 0.9 pp, high confidence |
| Mistral Medium 3 | 18.3% | 81.5% | estimated ± 0.9 pp, high confidence |
| Ling 3.0 Flash | 18.2% | 81.3% | estimated ± 0.9 pp, high confidence |
| Magistral Small | 18.0% | 81.1% | estimated ± 0.9 pp, high confidence |
| Claude 3 Haiku | 17.6% | 80.8% | estimated ± 0.9 pp, high confidence |
| A.X K2 | 17.5% | 80.7% | estimated ± 0.9 pp, high confidence |
| Qwen3.8-27B | 17.3% | 80.6% | estimated ± 0.9 pp, high confidence |
| Grok 4.1 Fast | 17.2% | 80.6% | estimated ± 0.9 pp, high confidence |
| Nova Pro | 16.9% | 80.5% | estimated ± 0.9 pp, high confidence |
| K-Exaone | 16.4% | 80.4% | estimated ± 0.9 pp, high confidence |
| GLM-4.5-Air | 16.3% | 80.4% | estimated ± 0.9 pp, high confidence |
| Solar Pro 2 | 16.1% | 80.4% | estimated ± 0.9 pp, high confidence |
| Gemma 4 12B | 15.7% | 80.4% | estimated ± 0.9 pp, high confidence |
| Ling 2.6 Flash | 15.6% | 80.4% | estimated ± 0.9 pp, high confidence |
| MiMo-V2-Flash | 15.6% | 80.4% | estimated ± 0.9 pp, high confidence |
| Gemini 2.5 Flash-Lite | 15.4% | 80.4% | estimated ± 0.9 pp, high confidence |
| Devstral Small | 14.7% | 80.3% | estimated ± 0.9 pp, high confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 14.7% | 80.3% | estimated ± 0.9 pp, high confidence |
| K-EXAONE 2.0 | 14.5% | 80.3% | estimated ± 0.9 pp, high confidence |
| Gemma 3 27B | 13.0% | 80.3% | estimated ± 0.9 pp, medium confidence |
| Granite 4.2 8B | 11.8% | 80.3% | estimated ± 0.9 pp, medium confidence |
| Granite 4.2 30B | 11.2% | 80.3% | estimated ± 0.9 pp, medium confidence |
| Command A+ | 9.8% | 80.3% | estimated ± 0.9 pp, medium confidence |
| Gemma 4 E4B | 9.7% | 80.3% | estimated ± 0.9 pp, medium confidence |
| LFM2.5-8B-A1B | 9.4% | 80.3% | estimated ± 0.9 pp, medium confidence |
| Granite 4.2 3B | 9.2% | 80.3% | estimated ± 0.9 pp, medium confidence |
| LFM3-1.7B | 7.1% | 80.3% | estimated ± 0.9 pp, medium confidence |
Fitted on
| Model | AA-Omniscience Accuracy | MMLU |
|---|---|---|
| Claude 3.5 Haiku | 13.2% | 80.3% |
| DeepSeek-R1 | 30.5% | 90.8% |
| GPT-4.1 | 27.8% | 90.2% |
| GPT-4.1 mini | 20.3% | 87.5% |
| GPT-4.1 nano | 13.7% | 80.1% |
| GPT-OSS 120B | 21.7% | 90.0% |
| Kimi K2 | 27.4% | 89.5% |
| Mistral Small 3.1 | 15.1% | 80.6% |
| o1 | 34.5% | 91.8% |
| o4-mini | 24.8% | 90.0% |