Calibration

WideResearch → GDPval-AA

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

GDPval-AA is estimated from WideResearch with a inverse Michaelis–Menten curve fitted on 6 models measured on both: y = 0.05066·(x − 0.0000) / (0.0000 + 0.9165 − x), R² = 0.74, cross-validated error 14.1 pp. It is used for 0 estimates.

Fitted on

Model	WideResearch	GDPval-AA
Kimi K2.6	80.8%	26.9%
Kimi K2.5	72.7%	17.0%
Ling 3.0 Flash	73.6%	22.4%
Ling 3.1 Flash	83.3%	55.9%
Qwen3.6-35B-A3B	60.1%	19.6%
Qwen3.6 Plus	74.3%	24.5%