MBZUAI/medix-rl-data

---
dataset_info:
features:
- name: id
dtype: string
- name: image
list:
image:
mode: RGBA
- name: source
dtype: string
- name: problem
dtype: string
- name: solution
dtype: string
splits:
- name: test
num_bytes: 1226261011
num_examples: 2451
- name: train
num_bytes: 22901597169
num_examples: 51335
download_size: 47058109860
dataset_size: 24127858180
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
- split: train
path: data/train-*
license: cc
task_categories:
- image-text-to-text
language:
- en
tags:
- medical
- biology
size_categories:
- 10K<n<100K
---

# MediX-R1: Open-Ended Medical Reinforcement Learning

<p align="center">
<img src="https://huggingface.co/MBZUAI/MediX-R1-30B/resolve/main/assets/logo_white_no_bg.png" alt="MediX-R1" width="200">
</p>

<p align="center">
<img src="https://i.imgur.com/waxVImv.png" alt="MediX-R1">
</p>

#### **Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI), UAE**

[](https://medix.cvmbzuai.com)
[](https://arxiv.org/pdf/2602.23363)
[](https://huggingface.co/collections/MBZUAI/medix-r1)
[](https://medix.cvmbzuai.com/leaderboard)

---

## Model Zoo

| Model | HuggingFace |
|-------|-------------|
| MediX-R1-2B | [MBZUAI/MediX-R1-2B](https://huggingface.co/MBZUAI/MediX-R1-2B) |
| MediX-R1-8B | [MBZUAI/MediX-R1-8B](https://huggingface.co/MBZUAI/MediX-R1-8B) |
| MediX-R1-30B | [MBZUAI/MediX-R1-30B](https://huggingface.co/MBZUAI/MediX-R1-30B) |

---

## Citation

If you use MediX-R1 in your research, please cite our work as follows:

```bibtex
@misc{mullappilly2026medixr1openendedmedical,
title={MediX-R1: Open Ended Medical Reinforcement Learning},
author={Sahal Shaji Mullappilly and Mohammed Irfan Kurpath and Omair Mohamed and Mohamed Zidan and Fahad Khan and Salman Khan and Rao Anwer and Hisham Cholakkal},
year={2026},
eprint={2602.23363},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2602.23363},
}
```

---

## License

This project is released for **research purposes only** under [*CC-BY-NC-SA 4.0*](https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode.en) License. It is not intended for clinical or commercial use.

Users are urged to employ MediX-R1 responsibly, especially when applying its outputs in real-world medical scenarios. It is imperative to verify the model's advice with qualified healthcare professionals and not rely on it for medical diagnoses or treatment decisions.

---

## Acknowledgements

We are thankful to [EasyR1](https://github.com/hiyouga/EasyR1) (a fork of [veRL](https://github.com/volcengine/verl)) for their open-source RL training framework.

This work was partially supported with *NVIDIA Academic Grant 2025* and *MBZUAI-IITD* Research Collaboration Seed Grant.

We are grateful to [MBZUAI](https://mbzuai.ac.ae/) for compute and support.