EVID-Bench - Selective Omission: leaderboard

Metric: Point-level accuracy (%) on the 25 selective omission videos (single-source editing: key context cut out), 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 March 2027. 9 models tracked.

Top models

#ModelScore
1GPT-5.572.92
2Claude Opus 4.669.79
3Gemini 3.1 Pro (Preview)67.7
4Claude Sonnet 4.666.67
5Gemini 3 Flash65.6
6GPT-5.464.58
7Qwen 3 VL 235B A22B Instruct61.5
8Qwen 3.5 Plus58.3
9GPT-5.4 Mini57.29

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