benchgap
Calibration

ApprenticeBench → DeepSearchQA

DeepSearchQA is estimated from ApprenticeBench with a linear curve fitted on 5 models measured on both: y = 0.7580·x + 0.7185, R² = 0.65, cross-validated error 7.5 pp. It is used for 2 estimates.

Estimated modelApprenticeBenchDeepSearchQASource
Claude Opus 4.77.0%77.2%estimated ± 7.5 pp, low confidence
Claude Sonnet 4.62.0%73.4%estimated ± 7.5 pp, low confidence