Calibration
ARC-AGI-3 → HLE
HLE is estimated from ARC-AGI-3 with a Michaelis–Menten curve fitted on 5 models measured on both: y = 0.6094·x / (0.00033 + x), R² = 0.57, cross-validated error 5.5 pp. It is used for 12 estimates.
| Estimated model | ARC-AGI-3 | HLE | Source |
|---|---|---|---|
| Gemini 3.1 Pro | 0.4% | 56.6% | estimated ± 5.5 pp, low confidence |
| Gemini 3.8 Flash | 10.4% | 60.7% | estimated ± 5.5 pp, low confidence |
| GPT-5.6 Luna | 0.2% | 51.6% | estimated ± 5.5 pp, low confidence |
| GPT-5.6 Sol | 7.8% | 60.7% | estimated ± 5.5 pp, low confidence |
| GPT-5.6 Terra | 0.8% | 58.6% | estimated ± 5.5 pp, low confidence |
| GPT-6.1 Sol | 52.7% | 60.9% | estimated ± 5.5 pp, low confidence |
| GPT-6 Astra | 62.7% | 60.9% | estimated ± 5.5 pp, low confidence |
| GPT-6 Luna | 0.1% | 46.0% | estimated ± 5.5 pp, low confidence |
| GPT-6 Sol | 4.6% | 60.5% | estimated ± 5.5 pp, low confidence |
| Grok 4.20 | 0.1% | 44.8% | estimated ± 5.5 pp, low confidence |
| Grok 4.5 | 0.3% | 55.0% | estimated ± 5.5 pp, low confidence |
| Grok 4.6 | 2.1% | 60.0% | estimated ± 5.5 pp, low confidence |