benchgap
Calibration

MCP Atlas → Gert Labs

Gert Labs is estimated from MCP Atlas with a linear curve fitted on 13 models measured on both: y = 0.3586·x + 0.3667, R² = 0.45, cross-validated error 8.2 pp. It is used for 3 estimates.

Estimated modelMCP AtlasGert LabsSource
Kimi K2.7 Code76.0%63.9%estimated ± 8.2 pp, low confidence
LLaDA2.2-flash46.2%53.2%estimated ± 8.2 pp, low confidence
Solar Open 258.2%57.5%estimated ± 8.2 pp, low confidence