Calibration

Vals GPQA Diamond → LABBench2

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

LABBench2 is estimated from Vals GPQA Diamond with a Michaelis–Menten + offset curve fitted on 6 models measured on both: y = 0.0000 + 2.0000·x / (1.31705 + x), R² = 0.43, cross-validated error 1.9 pp. It is used for 0 estimates.

Fitted on

Model	Vals GPQA Diamond	LABBench2
Claude Opus 5	93.4%	84.2%
Claude Sonnet 5	88.9%	80.1%
Gemini 3.7 Flash	93.9%	82.1%
Gemini 3.8 Flash	94.4%	86.2%
GPT-5.6 Sol	95.2%	82.1%
GPT-5.6 Terra	90.9%	81.2%