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
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 43.24 |
| 2 | Claude Opus 4.6 | 38.74 |
| 3 | Gemini 3.1 Pro (Preview) | 36.9 |
| 4 | GPT-5.4 | 35.59 |
| 5 | Gemini 3 Flash | 34.7 |
| 6 | Claude Sonnet 4.6 | 34.23 |
| 7 | Qwen 3.5 Plus | 34.2 |
| 8 | Qwen 3 VL 235B A22B Instruct | 33.3 |
| 9 | GPT-5.4 Mini | 26.13 |
Interactive version: theaggregate.ai/benchmark?slug=evid-bench-video-level · How It Works · Data refreshed daily, snapshot 2026-09-29.