AEGIS (Academic Image Forensics) - Textual Artifact Recognition: 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. Macro-F1 for deciding whether the text inside an image shows AI-synthesis traces, in percent; higher is better.
Metric: Macro-F1 (%) over two classes, telling whether the text regions of an image carry AI-synthesis traces, 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 February 2027. 26 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 84.67 | #119 | |
| 2 | GPT-4.1 | 82.74 | #505 | |
| 3 | Gemini 2.5 Flash | 80.97 | #463 | |
| 4 | O4 Mini | High | 80.02 | #368 |
| 5 | Qwen 3 VL Plus | 78.98 | #512 | |
| 6 | Doubao-Seed-1.6 | Thinking | 77.57 | |
| 7 | Seed-1.6 | 76.9 | #469 | |
| 8 | GPT-5.1 | 76.38 | #240 | |
| 9 | Doubao-1.5-thinking-vision-pro | 75.44 | #344 | |
| 10 | Qwen-VL-Max (AEGIS checkpoint unspecified) | 73.74 | ||
| 11 | Doubao-1.5 Vision Pro | 72.32 | #1026 | |
| 12 | Qwen 2.5 VL 72B Instruct | 70.76 | #841 | |
| 13 | Llama 4 Maverick | 68.32 | #1110 | |
| 14 | Ministral 3 14B | 62.25 | #1870 | |
| 15 | Doubao-Seed-1.6-Flash | 60.39 | #1250 |
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-textual-artifact-recognition · How It Works · Data refreshed daily, snapshot 2026-11-21.