AnalysisAI ModelsNovember 6, 2026

Hybrid Jev + Opus judge hits frontier accuracy at 4x lower cost

Read original source →hightouch.com

Hightouch found Jev and Luna decision models scored 88.8–97.3% out of the box on identity-resolution grading, trailing Opus by ~6 points on harder cases. Routing low-confidence cases to Opus matched frontier accuracy at over 4x lower cost; decision models run $0.5–$0.10 per 1,000 identities in ~0.15s.

1 source

More stories today

Open the live feed