Calibration

ExploitGym → Toolathlon

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

Toolathlon is estimated from ExploitGym with a Hill curve fitted on 6 models measured on both: y = 0.9869 + (0.5494 − 0.9869)·x^6.00 / (0.00785^6.00 + x^6.00), R² = 0.96, cross-validated error 9.0 pp. It is used for 1 estimate.

Estimated model	ExploitGym	Toolathlon	Source
Claude Mythos Preview	17.5%	54.9%	estimated ± 9.0 pp, low confidence