Calibration

ARC-AGI-3 → ARC-AGI-2

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

ARC-AGI-2 is estimated from ARC-AGI-3 with a Hill curve fitted on 17 models measured on both: y = 0.0000 + (0.9957 − 0.0000)·x^0.37 / (0.00031^0.37 + x^0.37), R² = 0.67, cross-validated error 9.4 pp. It is used for 0 estimates.