EVID-Bench - Object Manipulation: leaderboard
Metric: Point-level accuracy (%) on the 24 object manipulation videos (AI generation: a generated object edit), 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 | 13.89 |
| 2 | GPT-5.4 | 11.11 |
| 3 | Claude Opus 4.6 | 11.11 |
| 4 | Claude Sonnet 4.6 | 9.72 |
| 5 | Gemini 3.1 Pro (Preview) | 9.7 |
| 6 | Qwen 3.5 Plus | 9.7 |
| 7 | GPT-5.4 Mini | 5.56 |
| 8 | Qwen 3 VL 235B A22B Instruct | 4.2 |
| 9 | Gemini 3 Flash | 2.8 |
Interactive version: theaggregate.ai/benchmark?slug=evid-bench-object-manipulation · How It Works · Data refreshed daily, snapshot 2026-09-29.