EVID-Bench - Synthetic Insertion: leaderboard
EVID-Bench (2026) gives a multimodal model a manipulated video inside a fixed search-and-compare pipeline; it must find related source videos on the web and state what information is false. This board is point-level accuracy in percent on the 24 synthetic insertion videos (AI generation: a generated clip inserted into real footage); higher is better.
Metric: Point-level accuracy (%) on the 24 synthetic insertion videos (AI generation: a generated clip inserted into real footage), 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 | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 16.67 | #49 |
| 2 | Gemini 3.1 Pro (Preview) | 13.9 | #94 |
| 3 | Qwen 3.5 Plus | 13.9 | #232 |
| 4 | Claude Opus 4.6 | 13.89 | #102 |
| 5 | Claude Sonnet 4.6 | 12.5 | #161 |
| 6 | Qwen 3 VL 235B A22B Instruct | 12.5 | #522 |
| 7 | Gemini 3 Flash | 11.1 | #164 |
| 8 | GPT-5.4 | 9.72 | #123 |
| 9 | GPT-5.4 Mini | 6.94 | #389 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=evid-bench-synthetic-insertion · How It Works · Data refreshed daily, snapshot 2026-11-13.