Calibration

MCP Atlas → HLE w/ tools

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

HLE w/ tools is estimated from MCP Atlas with a inverse Michaelis–Menten curve fitted on 8 models measured on both: y = 0.95997·(x − 0.0641) / (0.0641 + 2.0000 − x), R² = 0.72, cross-validated error 7.2 pp. It is used for 2 estimates.

Estimated model	MCP Atlas	HLE w/ tools	Source
LLaDA2.2-flash	46.2%	23.9%	estimated ± 7.2 pp, low confidence
Solar Open 2	58.2%	33.5%	estimated ± 7.2 pp, medium confidence