MI-CXR: leaderboard
Longitudinal chest X-ray reasoning on MI-CXR (2026), five-option questions over five-visit chest X-ray timelines from MIMIC-CXR, answered zero-shot from all five images: temporal event localization, interval-wise change reasoning and trajectory summarization. Overall accuracy over 5,311 questions. Accuracy in percent (chance 20); higher is better.
Metric: Overall accuracy over the TEL, ICR and GTS task families, five-option multiple-choice accuracy (%, times 100; chance 20) on MI-CXR, longitudinal reasoning over five-visit chest X-ray timelines from MIMIC-CXR (5,311 questions), zero-shot single-step prompting with all five images, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 14 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3.5-38B | 41.8 | #1021 |
| 2 | GPT-5.2 | 41.1 | #185 |
| 3 | Gemini 3 Pro (Preview) | 38.7 | #119 |
| 4 | InternVL3.5-8B | 35.8 | #1604 |
| 5 | Claude Sonnet 4.5 | 31.5 | #261 |
| 6 | MedGemma-27B-IT | 29.9 | #1350 |
| 7 | InternVL3.5-14B | 28.1 | #1504 |
| 8 | Qwen 3 VL 32B | 27.2 | #622 |
| 9 | Lingshu-32B | 24.7 | #1113 |
| 10 | IDEFICS2-8B | 24.5 | #5013 |
| 11 | MedGemma-4B | 23.7 | #2557 |
| 12 | Qwen 3 VL 8B | 23.4 | #917 |
| 13 | Lingshu-7B | 22.3 | #1635 |
| 14 | MI-CXR DeepSeek-VL-16B (checkpoint unspecified) | 18.1 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=mi-cxr · How It Works · Data refreshed daily, snapshot 2026-11-21.