Calibration
Artificial Analysis Intelligence Index → HLE
HLE is estimated from Artificial Analysis Intelligence Index with a offset logistic curve fitted on 48 models measured on both: y = 0.1940 + (0.4762 − 0.1940) / (1 + exp(−74.43·(x − 0.2310))), R² = 0.57, cross-validated error 9.9 pp. It is used for 11 estimates.
| Estimated model | Artificial Analysis Intelligence Index | HLE | Source |
|---|---|---|---|
| Claude 3 Opus | 8.7% | 19.4% | estimated ± 9.9 pp, low confidence |
| Claude 4.1 Opus | 18.6% | 20.3% | estimated ± 9.9 pp, medium confidence |
| DeepSeek R1 Distill Qwen 32B | 8.4% | 19.4% | estimated ± 9.9 pp, low confidence |
| Gemini 1.0 Pro | 5.3% | 19.4% | estimated ± 9.9 pp, low confidence |
| Gemini 1.5 Pro | 7.9% | 19.4% | estimated ± 9.9 pp, low confidence |
| GPT-4 Turbo | 7.0% | 19.4% | estimated ± 9.9 pp, low confidence |
| GPT-4o mini | 6.7% | 19.4% | estimated ± 9.9 pp, low confidence |
| o1-preview | 11.4% | 19.4% | estimated ± 9.9 pp, low confidence |
| o3-pro | 21.9% | 27.5% | estimated ± 9.9 pp, medium confidence |
| Phi-4 Multimodal Instruct | 5.8% | 19.4% | estimated ± 9.9 pp, low confidence |
| Qwen2.5 Coder 32B Instruct | 6.7% | 19.4% | estimated ± 9.9 pp, low confidence |