AgentRx (Multimodal, Few-Shot) - Length of Stay: leaderboard
Clinical risk prediction by vision-language models on MIMIC-IV ICU admissions (AgentRx, Al Jorf and Shamout, 2026): from a patient summary of the history and demographics recorded before ICU admission plus, where available, the EHR time series and a chest X-ray from the first 48 ICU hours and the radiology reports, the model gives the probability that the stay lasts more than seven days. Single agent with a few-shot prompt holding one positive and one negative training example; AUROC on the test split, from 0 to 1, higher is better.
Metric: AUROC (0-1; stay beyond seven days; MIMIC-IV ICU test split; input: a patient summary of the history and demographics recorded before ICU admission plus, where available, the EHR time series and a chest X-ray from the first 48 ICU hours and the radiology reports; single agent, few-shot prompt with one positive and one negative training example, predicted probability). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 0.68 | #1842 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=agentrx-multimodal-few-shot-length-of-stay · How It Works · Data refreshed daily, snapshot 2026-11-12.