Rather
Rather
Evidence & attribution
Every result points back to its publisher. Check source availability, reuse terms, and when each integration last refreshed.
Published website leaderboard aggregates, with original metric scales and source model labels. Dates are source-listed metadata, not verified release or publication dates. Website footer declares CC BY-SA 4.0.
Published results; generation and scoring provenance varies by submission.
VBench 1.0 text-to-video results. Published scores include certified and uncertified submissions; see each result for generation and scoring teams. Space declares MIT; no videos, prompts or private submission files are redistributed.
Original and Verified results are separate, with image-to-video and multiframe conditioning kept apart. Submission methods, report links and reported uncertainty are retained. CC BY 4.0: https://creativecommons.org/licenses/by/4.0/
Leaderboard data reuse terms need clarification. Code is Apache-2.0; CSV coverage is unconfirmed.
We retain the published values, model variants, protocol, and source revision. Known records can refresh automatically after an initial review. New identities and unexpected changes are held for review. Rather’s own evaluations will be labeled separately when available.
Missing results are not zero scores. A recent source check does not mean a model was recently evaluated.