Calibration

V* → RealWorldQA

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

RealWorldQA is estimated from V* with a Michaelis–Menten curve fitted on 6 models measured on both: y = 1.0034·x / (0.18918 + x), R² = 0.53, cross-validated error 2.9 pp. It is used for 0 estimates.

Fitted on

Model	V*	RealWorldQA
Claude Opus 4.5	67.0%	77.0%
Gemini 3 Pro	88.0%	83.3%
GPT-5.2	75.9%	83.3%
LFM3-VL-2B	86.4%	79.3%
Qwen3.5 397B	95.8%	83.9%
Qwen3.6-27B	94.7%	84.1%