← All benchmarks

AIM Grounding

Professional Work

Zeta's private summary-faithfulness benchmark: each model summarizes the same 1,001 customer reports a marketer would read before a campaign decision, and a judge checks every material claim of each generated statement (entity, scope, metric, value, comparison and implication) against the report's data. The score is the share of a model's roughly 1,450 statements judged supported, reported as Zeta's mean judge score. Zeta runs it and supplied the results to Fireworks' Specialized Intelligence Index.

Top models (higher is better)