AEGIS (Academic Image Forensics) - Textual Artifact Recognition: leaderboard
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 January 2027. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 84.67 |
| 2 | GPT-4.1 | 82.74 |
| 3 | Gemini 2.5 Flash | 80.97 |
| 4 | O4 Mini (High) | 80.02 |
| 5 | Doubao-Seed-1.6 (Thinking) | 77.57 |
| 6 | Seed-1.6 | 76.9 |
| 7 | GPT-5.1 | 76.38 |
| 8 | Qwen 2.5 VL 72B Instruct | 70.76 |
| 9 | Llama 4 Maverick | 68.32 |
| 10 | Ministral 3 14B | 62.25 |
| 11 | Doubao-Seed-1.6-Flash | 60.39 |
| 12 | Claude Sonnet 4.5 | 58.7 |
| 13 | Gemma 3 27B (IT) | 58.39 |
Interactive version: theaggregate.ai/benchmark?slug=aegis-academic-image-forensics-textual-artifact-recognition · How It Works · Data refreshed daily, snapshot 2026-10-07.