Calibration
ARC-AGI-1 → HLE
HLE is estimated from ARC-AGI-1 with a offset logistic curve fitted on 16 models measured on both: y = 0.4212 + (0.6476 − 0.4212) / (1 + exp(−60.69·(x − 0.9413))), R² = 0.67, cross-validated error 7.0 pp. It is used for 1 estimate.
| Estimated model | ARC-AGI-1 | HLE | Source |
|---|---|---|---|
| GPT-5.1 | 72.8% | 42.1% | estimated ± 7.0 pp, medium confidence |
Fitted on
| Model | ARC-AGI-1 | HLE |
|---|---|---|
| Claude Fable 5.1 | 97.5% | 65.0% |
| Claude Opus 4.8 | 92.5% | 57.9% |
| Claude Opus 5 | 97.5% | 64.7% |
| Claude Sonnet 4.6 | 86.0% | 49.0% |
| DeepSeek V4 Flash 0731 | 89.0% | 34.8% |
| DeepSeek V4 Pro | 90.0% | 42.7% |
| Gemini 3.5 Flash | 92.5% | 40.2% |
| GPT-5.4 | 93.7% | 52.1% |
| GPT-5.4 mini | 63.7% | 41.5% |
| GPT-5.4 nano | 51.5% | 37.7% |
| GPT-5.4 Pro | 94.5% | 58.7% |
| GPT-5.5 | 95.0% | 52.2% |
| GPT-5.5 Pro | 95.0% | 57.2% |
| Inkling 2 Small | 96.5% | 53.2% |
| Inkling-Small | 84.0% | 47.8% |
| Kimi K3 | 94.5% | 56.0% |