Meditron
EPFL's fully open medical LLM suite, now a framework for medicalising open base models
Overview
Meditron is a suite of open medical large language models from EPFL's LiGHT laboratory, released in November 2023 in 7B and 70B sizes built on Llama-2 and adapted on a curated medical corpus. In 2026 the team published MeditronFO, a fully open framework used to medicalise open base models including OLMo, EuroLLM and Switzerland's Apertus.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 4 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with Meditron
- Download open medical model weights in 7B and 70B sizes and run them on your own hardware
- Read the exact composition of the training corpus rather than accepting an undisclosed data mix
- Continue pretraining a base model on medical text using the published GAP-Replay recipe
- Reproduce the published medical benchmark evaluations instead of taking vendor scores on trust
- Use the MeditronFO framework to medicalise a fully open base model such as OLMo, EuroLLM or Apertus
- Deploy inside a hospital network with no inference traffic leaving the institution
Judge Meditron on transparency, not leaderboard position. The training corpus is documented, the code is open, and the 2026 MeditronFO release gives you the method rather than just the artefact — which is what you need if a reviewer will ask how a clinical model was built.
It is a research release and behaves like one: no support, no hosted endpoint, no clearance, and real infrastructure work between the download and anything a clinician should use. Strong choice for institutions that cannot send patient data off-site or that must reproduce their own model; wrong choice for a team that wants to buy something finished.
What people say
Meditron is the reference point people reach for when they want to argue that open medical models are viable, and the reason is provenance rather than raw capability. EPFL published the 7B and 70B models in November 2023, built on Llama-2 and adapted on a corpus the team documented in detail: clinical practice guidelines gathered from hospitals and international health bodies, PubMed abstracts, and full-text papers. Anyone can read what went in. That is unusual enough in medical AI to be the headline feature.
What has changed since is more interesting than a version bump. In 2026 the LiGHT laboratory released MeditronFO, a fully open framework for building medical models rather than a single model, and used it to medicalise open base models including OLMo, EuroLLM and Apertus, the Swiss model developed with ETH Zurich. The shift is from shipping weights to shipping the method, which matters if your constraint is that you must be able to explain and reproduce how a clinical model was made.
The practical caveats are the ones that attach to any research release. There is no support contract, no hosted endpoint, and no regulatory clearance — Meditron is a research artefact, and the benchmark results that circulate are exam-style evaluations rather than evidence of safe clinical use. Running the 70B model well needs real GPU capacity, and teams without an ML engineer will find the gap between downloading weights and deploying something a clinician can rely on considerably wider than the download suggests.
Where it earns its place is in settings that cannot send patient data to a commercial API at all, and in research groups who need to show their working. If you want a supported product, this is the wrong shelf. If you want a medical model whose training data and method you can inspect and rebuild, there is not much else at this level.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Meditron's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Meditron unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Meditron directly →
Spotted something out of date? Suggest an update →
Alternatives to Meditron
More in Healthcare
