Calibration

V* → CountBench

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

CountBench is estimated from V* with a offset logistic curve fitted on 5 models measured on both: y = 0.9056 + (0.9750 − 0.9056) / (1 + exp(−41.18·(x − 0.7937))), R² = 1.00, cross-validated error 0.9 pp. It is used for 1 estimate.

Estimated model	V*	CountBench	Source
Step 3.7 Flash	95.3%	97.5%	estimated ± 0.9 pp, medium confidence