AEGIS (Academic Image Forensics): leaderboard
AEGIS (2026) tests whether multimodal models can audit AI-generated or AI-edited images in academic papers, from figure panels across seven scientific image categories and forgeries made with 25 image generators. Composite forensic index that combines the three classification tasks with tampering localization and penalizes over-localization, in percent; higher is better.
Metric: Normalized Forensic Index (%): harmonic mean of the macro-F1 scores of the three classification tasks and the correct localization accuracy of tampering pinpointing, times the square root of one minus the over-localization rate, on AEGIS academic images (seven categories, 39 subtypes; real panels and forgeries from 25 generative models under four forgery strategies), zero-shot minimal prompts, lossless PNG input; higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 26 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-5.1 | 48.8 | #240 | |
| 2 | Gemini 2.5 Flash | 47.02 | #463 | |
| 3 | Doubao-Seed-1.6 | Thinking | 46.66 | |
| 4 | Gemini 3 Pro (Preview) | 45.79 | #119 | |
| 5 | GPT-4.1 | 43.31 | #505 | |
| 6 | Qwen-VL-Max (AEGIS checkpoint unspecified) | 43.01 | ||
| 7 | O4 Mini | High | 42.77 | #368 |
| 8 | Doubao-1.5-thinking-vision-pro | 42.39 | #344 | |
| 9 | Seed-1.6 | 41.73 | #469 | |
| 10 | Qwen 2.5 VL 72B Instruct | 38.71 | #841 | |
| 11 | Gemma 3 27B (IT) | 37.55 | #1324 | |
| 12 | Doubao-Seed-1.6-Flash | 36.03 | #1250 | |
| 13 | Doubao-1.5 Vision Pro | 34.07 | #1026 | |
| 14 | Qwen-VL-Plus (AEGIS checkpoint unspecified) | 33.89 | ||
| 15 | LLaVA-OneVision-72B | 33.81 | #1771 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=aegis-academic-image-forensics · How It Works · Data refreshed daily, snapshot 2026-11-21.