Calibration
ARC-AGI-3 → GPQA Diamond
GPQA Diamond is estimated from ARC-AGI-3 with a linear curve fitted on 14 models measured on both: y = 0.0616·x + 0.9255, R² = 0.47, cross-validated error 1.8 pp. It is used for 1 estimate.
| Estimated model | ARC-AGI-3 | GPQA Diamond | Source |
|---|---|---|---|
| Claude Fable 5.2 | 65.7% | 96.6% | estimated ± 1.8 pp, medium confidence |
Fitted on
| Model | ARC-AGI-3 | GPQA Diamond |
|---|---|---|
| Claude Opus 4.6 | 0.5% | 89.2% |
| Claude Opus 4.7 | 0.2% | 94.2% |
| Claude Opus 4.8 | 1.5% | 93.6% |
| Gemini 3.1 Pro | 0.4% | 94.3% |
| GPT-5.4 | 0.2% | 92.8% |
| GPT-5.5 | 0.4% | 93.6% |
| GPT-5.6 Luna | 0.2% | 92.3% |
| GPT-5.6 Sol | 7.8% | 94.6% |
| GPT-5.6 Terra | 0.8% | 92.9% |
| GPT-6.2 Astra | 70.3% | 97.0% |
| GPT-6.2 Sol | 54.9% | 96.0% |
| GPT-6 Astra | 62.7% | 96.0% |
| Grok 4.20 | 0.1% | 88.5% |
| Inkling 2 Small | 0.3% | 93.0% |