LiGHT, Laboratory for intelligent Global Health and Humanitarian Response Technologies

Meditron

Open medical AI models, built and tested with clinicians around the world.

Talk to MeditronJoin us

Where clinicians evaluated Meditron

    Clinical experts evaluate AI against real-world local contexts.

    Clinicians contribute realistic scenarios from their daily practice and evaluate blinded AI responses for clinical quality, safety and contextual relevance.

    Explore MOOVE

    What is Meditron

    Meditron is a family of open models for medicine: large language models trained to answer clinical questions. Anyone can download them, run them on their own hardware and adapt them, from a hospital in Switzerland to a health post in Malawi.

    1. 2023 Meditron 1 Open weights

      7B and 70B models based on Llama-2, pretrained on PubMed and clinical guidelines.

    2. 2024 Clinicians step in

      Sixteen physicians grade Meditron's answers and provide their first clinician analysis.

    3. 2025 Meditron 3 Open weights

      Co-designed with clinicians, 8B and 70B models based on Llama-3.1.

    4. 2026 MeditronFO Fully open

      Open data, open weights and open science, models based on Apertus-1 and four other open bases.

    Meet our latest project

    MeditronFO is the first fully-open pipeline for creating medical LLMs. On HealthBench, Apertus-70B-MeditronFO reaches 49.3%, up from 39.4% for its base; MedGemma-27B reaches 54.8%.

    HealthBench full

    Score vs model size
    5,000 conversations graded by Gemma-4-31B against physician-written rubrics. Arrows go from each base to its MeditronFO finetune; bars are 95% intervals.
    MeditronFO: An Auditable Pipeline for Clinical LLMsThe models, their data, the results and the paper.Project page

    Open data, open code, open models

    Among existing medical LLMs, the MeditronFO models are the only ones built on a fully-open base that release their medical data, the data pipeline and the training code.

    ModelFully-open baseMedical dataData pipelineTraining code
    Med-GeminiNoNoNoNo
    Meditron-3 (Llama 3.1)NoNoNoNo
    MedGemma-27BNoNoNoNo
    BioMistralNoYesNoNo
    Aloe BetaNoYesNoNo
    HuatuoGPT-o1NoYesNoYes
    Gemma-3-27B-MeditronFONoYesYesYes
    Apertus, OLMo-2, EuroLLM MeditronFOYesYesYesYes
    The MeditronFO collectionThe models and their training data, free to download.Hugging Face

    Talk to Meditron

    Hello, I'm Meditron, an open medical AI model. Ask me a medical question.

    I'm a research model, not a doctor: I can be wrong, and I know nothing about you beyond what you type.

    Meditron can be wrong. It is a research model, not a doctor. Use with care.

    Can you judge like a clinician?

    Five real MOOVE cases written by clinicians in Ethiopia, Switzerland and Kenya: read two anonymous answers, vote for the better one, and see whether you agree with the clinicians and with AutoMOOVE, our AI judge.

    AutoMOOVE: judge like a clinicianFive cases, two anonymous answers each.Game

    Join us

    Meditron is built with the people who use it.

    Clinicians

    Put tough questions from your own practice to the models at a MOOVE event, compare two anonymous answers and vote for the better one.

    Join MOOVE

    Community

    Everyone else: researchers, engineers, health workers or simply curious. Stay up to date with Meditron and join the community: fill in a short form and we send you an invite to our Slack.

    Join the community

    Cite Meditron

    Theimer-Lienhard, X., El-Amin, M., Elhassan, F., Vaidya, S., Cartier-Negadi, V., Sasu, D., Klein, L., & Hartley, M.-A. (2026). Fully Open Meditron: An Auditable Pipeline for Clinical LLMs. arXiv:2605.16215.

    @misc{theimerlienhard2026fullyopenmeditronauditable,
      title={Fully Open Meditron: An Auditable Pipeline for Clinical LLMs},
      author={Xavier Theimer-Lienhard and Mushtaha El-Amin and Fay Elhassan and Sahaj Vaidya and Victor Cartier-Negadi and David Sasu and Lars Klein and Mary-Anne Hartley},
      year={2026},
      eprint={2605.16215},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2605.16215}
    }

    Contact

    Questions about Meditron? Write to admin-light@epfl.ch.

    Meet the Meditron team

    Credits

    Built at EPFL's LiGHT lab with the Swiss AI Initiative and compute from CSCS (Alps). Thanks to the clinicians of the MOOVE initiative and the LiGHT physician review panel.

    • EPFL AI Center
    • ETH AI Center
    • CSCS, Swiss National Supercomputing Centre