Atlas

Benchmarks

← All benchmarks

Harvey LAB-AA - release-reported metric unspecified

Professional Work

Google's LAB-AA headline without an identified all-pass versus mean-criterion metric or AA version. Diagnostic; a score near90 is not treated as the newer hallucination-gated task all-pass percentage.

Top models (higher is better)

ModelScore
Gemini 3.7 Flash90.7
Loading Atlas data…