EVID-Bench - Video-Level: leaderboard

Metric: Video-level accuracy (%) over all 222 manipulated videos: a video counts only when every ground-truth misinformation point is identified, using the retrieval-augmented verification pipeline of the paper (chain-of-thought analysis, up to six YouTube search-verify-reflect rounds, 64 frames at 480p, temperature 0); each video has 3 to 5 annotated misinformation points, matched by a majority of three LLM judges; mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 9 models tracked.

Top models

#ModelScore
1GPT-5.543.24
2Claude Opus 4.638.74
3Gemini 3.1 Pro (Preview)36.9
4GPT-5.435.59
5Gemini 3 Flash34.7
6Claude Sonnet 4.634.23
7Qwen 3.5 Plus34.2
8Qwen 3 VL 235B A22B Instruct33.3
9GPT-5.4 Mini26.13

Interactive version: theaggregate.ai/benchmark?slug=evid-bench-video-level · How It Works · Data refreshed daily, snapshot 2026-09-29.