benchgap
Calibration

MCP Atlas → APEX-Agents-AA

APEX-Agents-AA is estimated from MCP Atlas with a linear curve fitted on 13 models measured on both: y = 0.4758·x + -0.0158, R² = 0.74, cross-validated error 5.7 pp. It is used for 7 estimates.

Estimated modelMCP AtlasAPEX-Agents-AASource
Kimi K2.7 Code76.0%34.6%estimated ± 5.7 pp, medium confidence
LLaDA2.2-flash46.2%20.4%estimated ± 5.7 pp, medium confidence
Muse Glimmer 30B75.5%34.3%estimated ± 5.7 pp, medium confidence
Qwen3.6-35B-A3B62.8%28.3%estimated ± 5.7 pp, medium confidence
Qwen3.6 Plus48.2%21.4%estimated ± 5.7 pp, medium confidence
Qwen3.7 Max76.4%34.8%estimated ± 5.7 pp, medium confidence
Solar Open 258.2%26.1%estimated ± 5.7 pp, medium confidence