benchgap
Calibration

AA ITBench → APEX-Agents-AA

APEX-Agents-AA is estimated from AA ITBench with a offset logistic curve fitted on 5 models measured on both: y = 0.3366 + (0.3939 − 0.3366) / (1 + exp(−200.00·(x − 0.4533))), R² = 0.80, cross-validated error 3.2 pp. It is used for 4 estimates.

Estimated modelAA ITBenchAPEX-Agents-AASource
DeepSeek V4.1 Flash46.9%39.2%estimated ± 3.2 pp, medium confidence
GLM-5.346.1%38.4%estimated ± 3.2 pp, medium confidence
GLM-5.3-Flash51.2%39.4%estimated ± 3.2 pp, medium confidence
Grok 4.742.1%33.7%estimated ± 3.2 pp, low confidence