Calibration

APEX-Agents → GDPval-AA

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

GDPval-AA is estimated from APEX-Agents with a Michaelis–Menten curve fitted on 6 models measured on both: y = 1.0123·x / (0.45552 + x), R² = 0.62, cross-validated error 8.0 pp. It is used for 2 estimates.

Estimated model	APEX-Agents	GDPval-AA	Source
Ling 3.0 Flash Fin	29.2%	39.5%	estimated ± 8.0 pp, low confidence
Solar Open 2	16.6%	27.0%	estimated ± 8.0 pp, low confidence