Calibration
Artificial Analysis Intelligence Index → MMLU-Redux
MMLU-Redux is estimated from Artificial Analysis Intelligence Index with a Hill curve fitted on 7 models measured on both: y = 0.7016 + (0.9521 − 0.7016)·x^6.00 / (0.11805^6.00 + x^6.00), R² = 0.95, cross-validated error 2.1 pp. It is used for 14 estimates.
| Estimated model | Artificial Analysis Intelligence Index | MMLU-Redux | Source |
|---|---|---|---|
| Claude 3 Opus | 8.7% | 73.7% | estimated ± 2.1 pp, low confidence |
| Claude 4.1 Opus | 18.6% | 93.7% | estimated ± 2.1 pp, medium confidence |
| Claude 4.1 Opus Thinking | 22.9% | 94.7% | estimated ± 2.1 pp, medium confidence |
| Claude Haiku 5.5 | 43.4% | 95.2% | estimated ± 2.1 pp, low confidence |
| DeepSeek R1 Distill Qwen 32B | 8.4% | 73.0% | estimated ± 2.1 pp, low confidence |
| Gemini 1.0 Pro | 5.3% | 70.4% | estimated ± 2.1 pp, low confidence |
| Gemini 1.5 Pro | 7.9% | 72.2% | estimated ± 2.1 pp, low confidence |
| GLM-5.3-Flash | 41.8% | 95.2% | estimated ± 2.1 pp, low confidence |
| GPT-4 Turbo | 7.0% | 71.2% | estimated ± 2.1 pp, low confidence |
| GPT-4o mini | 6.7% | 70.9% | estimated ± 2.1 pp, low confidence |
| o1-preview | 11.4% | 81.3% | estimated ± 2.1 pp, low confidence |
| o3-pro | 21.9% | 94.6% | estimated ± 2.1 pp, medium confidence |
| Phi-4 Multimodal Instruct | 5.8% | 70.5% | estimated ± 2.1 pp, low confidence |
| Qwen2.5 Coder 32B Instruct | 6.7% | 71.0% | estimated ± 2.1 pp, low confidence |