benchgap
Calibration

ARC-AGI-1 → HLE

HLE is estimated from ARC-AGI-1 with a offset logistic curve fitted on 15 models measured on both: y = 0.4220 + (0.6719 − 0.4220) / (1 + exp(−65.63·(x − 0.9434))), R² = 0.72, cross-validated error 6.4 pp. It is used for 1 estimate.

Estimated modelARC-AGI-1HLESource
GPT-5.172.8%42.2%estimated ± 6.4 pp, medium confidence