ToneBench · Style and scripts
Towards AI · Writes video scripts in an author's specified style, assessing openings, structure, expression, and spoken pacing.
What it measures and how
Uses 10 real script tasks, repeated generations, and three judges. Tasks, rubrics, judges, and run fingerprints must match. Hybrid fallback rows are excluded; configurations are not chosen by per-column maximum scores.
How this evidence is used
Under review: data access and current coverage are promising. Redisplay terms and pure-model configurations need verification.
Evaluation limits & attribution
Favors one Towards AI channel's style; exact tasks and reference scripts are unpublished. The current leader combines Fable 5 and Opus 4.8 fallback and cannot represent one model.
Data license: Public JSON is readable, but explicit aggregation and redisplay terms were not found.
Results are published by Towards AI. Original scores and the NexGateHub consensus index use different scales and cannot be added directly.