Calibration
AA-Omniscience Accuracy → SuperGPQA
SuperGPQA is estimated from AA-Omniscience Accuracy with a Hill curve fitted on 12 models measured on both: y = 0.6435 + (1.2000 − 0.6435)·x^6.00 / (0.45938^6.00 + x^6.00), R² = 0.59, cross-validated error 9.3 pp. It is used for 84 estimates.
| Estimated model | AA-Omniscience Accuracy | SuperGPQA | Source |
|---|---|---|---|
| Claude Fable 5 | 65.4% | 100.0% | estimated ± 9.3 pp, low confidence |
| Claude Opus 5.5 | 65.5% | 100.0% | estimated ± 9.3 pp, low confidence |
| Claude Sonnet 5.5 | 51.5% | 100.0% | estimated ± 9.3 pp, low confidence |
| Gemini 3.7 Flash | 55.3% | 100.0% | estimated ± 9.3 pp, low confidence |
| Gemini 3 Pro | 55.8% | 100.0% | estimated ± 9.3 pp, low confidence |
| Gemini 4 Carbon | 58.3% | 100.0% | estimated ± 9.3 pp, low confidence |
| GPT-6.1 Sol | 61.8% | 100.0% | estimated ± 9.3 pp, low confidence |
| GPT-6 Sol | 54.5% | 100.0% | estimated ± 9.3 pp, low confidence |
| Grok 4.5 | 51.6% | 100.0% | estimated ± 9.3 pp, low confidence |
| Gemini 3.8 Flash | 50.5% | 99.9% | estimated ± 9.3 pp, low confidence |
| Gemini 4 Hydrogen | 50.5% | 99.9% | estimated ± 9.3 pp, low confidence |
| Gemini 3.6 Flash | 50.0% | 99.1% | estimated ± 9.3 pp, low confidence |
| Gemini 4 Argon | 48.3% | 96.3% | estimated ± 9.3 pp, low confidence |
| Grok 4.6 | 48.2% | 96.2% | estimated ± 9.3 pp, low confidence |
| GPT-6.1 Terra | 47.6% | 95.1% | estimated ± 9.3 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 47.0% | 94.1% | estimated ± 9.3 pp, low confidence |
| Claude Opus 4.5 Thinking | 46.6% | 93.4% | estimated ± 9.3 pp, low confidence |
| Gemini 3 Flash | 45.8% | 91.9% | estimated ± 9.3 pp, medium confidence |
| Muse Spark 1.2 | 45.4% | 91.2% | estimated ± 9.3 pp, medium confidence |
| MiniMax M3.1 Flash | 45.1% | 90.6% | estimated ± 9.3 pp, medium confidence |
| MiniMax M3 | 44.9% | 90.3% | estimated ± 9.3 pp, medium confidence |
| Muse Spark 1.3 | 43.6% | 87.8% | estimated ± 9.3 pp, medium confidence |
| Grok 4.7 | 42.7% | 86.2% | estimated ± 9.3 pp, medium confidence |
| Muse Spark Open | 42.0% | 84.9% | estimated ± 9.3 pp, medium confidence |
| GPT-5.2-Codex | 41.1% | 83.2% | estimated ± 9.3 pp, medium confidence |
| Muse Wonder Full | 41.0% | 83.0% | estimated ± 9.3 pp, medium confidence |
| Grok 4 | 40.5% | 82.1% | estimated ± 9.3 pp, medium confidence |
| GPT-5 | 40.3% | 81.8% | estimated ± 9.3 pp, medium confidence |
| GPT-5.1-Codex | 39.9% | 81.1% | estimated ± 9.3 pp, medium confidence |
| GPT-5.1-Codex-Max | 39.9% | 81.1% | estimated ± 9.3 pp, medium confidence |
| GPT-6 Luna | 39.6% | 80.5% | estimated ± 9.3 pp, medium confidence |
| Muse Wonder | 38.0% | 77.8% | estimated ± 9.3 pp, medium confidence |
| Ling 3.0 Tiny | 37.9% | 77.7% | estimated ± 9.3 pp, medium confidence |
| GPT-5.1 | 37.7% | 77.4% | estimated ± 9.3 pp, medium confidence |
| MiMo-V2.6-Pro | 34.8% | 73.2% | estimated ± 9.3 pp, medium confidence |
| Claude Haiku 5.5 | 34.7% | 73.1% | estimated ± 9.3 pp, medium confidence |
| GLM-5.3 | 32.3% | 70.3% | estimated ± 9.3 pp, medium confidence |
| Beam 1.0 501B-A23B | 30.7% | 68.9% | estimated ± 9.3 pp, medium confidence |
| Qwen3.8 Max Preview | 29.5% | 68.0% | estimated ± 9.3 pp, medium confidence |
| GLM-5V-Turbo | 29.3% | 67.9% | estimated ± 9.3 pp, medium confidence |
| GLM-5-Turbo | 28.4% | 67.3% | estimated ± 9.3 pp, medium confidence |
| Apodex 1.1 Mini | 28.2% | 67.2% | estimated ± 9.3 pp, medium confidence |
| Muse Glimmer 30B | 28.0% | 67.1% | estimated ± 9.3 pp, medium confidence |
| MiMo-V2.6-Flash | 27.2% | 66.6% | estimated ± 9.3 pp, medium confidence |
| MiMo-V2-Pro | 26.6% | 66.4% | estimated ± 9.3 pp, medium confidence |
| Gemini 3.5 Flash-Lite | 26.5% | 66.3% | estimated ± 9.3 pp, medium confidence |
| Gemini 2.5 Flash | 26.1% | 66.2% | estimated ± 9.3 pp, medium confidence |
| Mistral Large 4 | 26.0% | 66.1% | estimated ± 9.3 pp, medium confidence |
| Step 3.7 Flash | 25.8% | 66.0% | estimated ± 9.3 pp, medium confidence |
| Grok 4.1 Fast (Reasoning) | 25.1% | 65.8% | estimated ± 9.3 pp, medium confidence |
| Mistral Large 3 | 25.0% | 65.8% | estimated ± 9.3 pp, medium confidence |
| Qwen3 Max | 24.4% | 65.6% | estimated ± 9.3 pp, medium confidence |
| DeepSeek V3 0324 | 24.3% | 65.5% | estimated ± 9.3 pp, medium confidence |
| DeepSeek V3.2 | 24.0% | 65.5% | estimated ± 9.3 pp, medium confidence |
| Grok Code Fast 1 | 23.5% | 65.3% | estimated ± 9.3 pp, medium confidence |
| Mistral Medium 3.5 128B | 23.0% | 65.2% | estimated ± 9.3 pp, medium confidence |
| Grok 4 Fast (Reasoning) | 22.8% | 65.2% | estimated ± 9.3 pp, medium confidence |
| Claude 4 Sonnet | 22.7% | 65.1% | estimated ± 9.3 pp, medium confidence |
| Mercury 2.5 | 22.0% | 65.0% | estimated ± 9.3 pp, medium confidence |
| Nemotron 3.5 Bolt 74B-A9B BF16 | 21.8% | 65.0% | estimated ± 9.3 pp, medium confidence |
| GLM-4.6 | 21.4% | 64.9% | estimated ± 9.3 pp, medium confidence |
| Mistral Small 4 | 21.4% | 64.9% | estimated ± 9.3 pp, medium confidence |
| Muse Ember | 21.1% | 64.9% | estimated ± 9.3 pp, medium confidence |
| GPT-4o | 19.9% | 64.7% | estimated ± 9.3 pp, medium confidence |
| Mistral Large 2 | 19.9% | 64.7% | estimated ± 9.3 pp, medium confidence |
| MiMo-V2-Omni | 19.3% | 64.7% | estimated ± 9.3 pp, low confidence |
| Solar Pro 3 | 18.5% | 64.6% | estimated ± 9.3 pp, low confidence |
| North Mini Code | 18.4% | 64.6% | estimated ± 9.3 pp, low confidence |
| Solar Pro 4 | 18.4% | 64.6% | estimated ± 9.3 pp, low confidence |
| Mistral Medium 3 | 18.3% | 64.6% | estimated ± 9.3 pp, low confidence |
| Claude 3 Haiku | 17.6% | 64.5% | estimated ± 9.3 pp, low confidence |
| Grok 4.1 Fast | 17.2% | 64.5% | estimated ± 9.3 pp, low confidence |
| Nova Pro | 16.9% | 64.5% | estimated ± 9.3 pp, low confidence |
| K-Exaone | 16.4% | 64.5% | estimated ± 9.3 pp, low confidence |
| GLM-4.5-Air | 16.3% | 64.5% | estimated ± 9.3 pp, low confidence |
| Solar Pro 2 | 16.1% | 64.4% | estimated ± 9.3 pp, low confidence |
| Mistral Small 3.1 | 15.1% | 64.4% | estimated ± 9.3 pp, low confidence |
| Devstral Small | 14.7% | 64.4% | estimated ± 9.3 pp, low confidence |
| GPT-4.1 nano | 13.7% | 64.4% | estimated ± 9.3 pp, low confidence |
| Claude 3.5 Haiku | 13.2% | 64.4% | estimated ± 9.3 pp, low confidence |
| Gemma 3 27B | 13.0% | 64.4% | estimated ± 9.3 pp, low confidence |
| Command A+ | 9.8% | 64.4% | estimated ± 9.3 pp, low confidence |
| LFM2.5-8B-A1B | 9.4% | 64.4% | estimated ± 9.3 pp, low confidence |
| Granite 4.2 3B | 9.2% | 64.3% | estimated ± 9.3 pp, low confidence |
Fitted on
| Model | AA-Omniscience Accuracy | SuperGPQA |
|---|---|---|
| Claude Opus 4.5 | 40.9% | 70.6% |
| Claude Opus 4.6 | 45.8% | 95.0% |
| Claude Sonnet 4.6 | 38.6% | 95.0% |
| GLM-5 | 26.3% | 66.8% |
| Kimi K2 | 27.4% | 57.2% |
| Kimi K2.5 | 35.2% | 69.2% |
| Qwen3.5-27B | 20.7% | 65.6% |
| Qwen3.5-35B-A3B | 20.1% | 63.4% |
| Qwen3.6-27B | 19.6% | 66.0% |
| Qwen 3.6 Max (preview) | 37.9% | 73.9% |
| Qwen3.6 Plus | 26.4% | 71.6% |
| Qwen3.7 Max | 31.1% | 73.6% |