GazeVaLM - Visual Turing Test: leaderboard
GazeVaLM (2026) pairs radiologist eye-tracking data with a visual Turing test on chest X-rays, half real and half generated by a diffusion model from the same reports. This board is accuracy in percent at telling real from synthetic images, one image per new chat session; higher is better.
Metric: Accuracy (%) classifying each of 60 chest X-rays (30 real MIMIC-CXR images, 30 RoentGen diffusion images generated from the same reports) as real or synthetic, a new chat session per image; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-5.2 | Thinking | 71.67 | #185 |
| 2 | Gemini 3 Pro | 66.67 | #145 | |
| 3 | Claude Sonnet 4.5 | Thinking | 65 | #261 |
| 4 | Grok 4.1 | 58.33 | #462 | |
| 5 | Llama 4 (GazeVaLM checkpoint unspecified) | 55 | ||
| 6 | DeepSeek V3.2 | 50 | #419 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=gazevalm-visual-turing-test · How It Works · Data refreshed daily, snapshot 2026-11-21.