FAQ (Deepfake Forensics) - Facial Perception: leaderboard
Metric: Accuracy (%) on FAQ's Level 1 facial-perception questions: four-option multiple choice on whether named facial regions or face boundaries in a FaceForensics++ deepfake or authentic video look clear or blurred (region and edge perception; LLM-written distractors, human-verified answers), zero-shot with Top-1 option matching; chance 25 percent; higher is better. Source: arxiv.org. Saturation forecast: Around March 2028. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 40 | #237 |
| 2 | GPT-4o | 26.9 | #333 |
| 3 | Qwen 2.5 VL 7B Instruct | 24.1 | #643 |
| 4 | InternVL2-8B | 24 | #826 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=faq-deepfake-forensics-facial-perception · How It Works · Data refreshed daily, snapshot 2026-10-11.