EVID-Bench - Contextual Fabrication: leaderboard

Metric: Point-level accuracy (%) on the 25 contextual fabrication videos (multi-source editing: footage placed in a false context), 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 December 2026. 9 models tracked.

Top models

#ModelScore
1GPT-5.588.46
2Claude Opus 4.686.54
3Claude Sonnet 4.679.81
4GPT-5.478.85
5Gemini 3.1 Pro (Preview)77.9
6Gemini 3 Flash75
7Qwen 3.5 Plus74
8Qwen 3 VL 235B A22B Instruct71.2
9GPT-5.4 Mini67.31

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