Med42
Abu Dhabi's open-access clinical LLMs, fine-tuned from Llama 3 and free to download
Overview
Med42 is a family of open-access clinical language models from M42, the Abu Dhabi health group. The first version, a Llama 2 70B fine-tune trained on Cerebras hardware with Core42, appeared in October 2023. Med42-v2, released in 2024 in 8B and 70B sizes, is built on Llama 3 and published on Hugging Face under the Llama 3 Community License.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Oct 2026 and may be out of date or incomplete.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 Med42
- Med42-v2 clinical model weights in 8B and 70B sizes on Hugging Face
- Clinical question answering run entirely inside your own infrastructure
- Published MedQA, MedMCQA, USMLE and MMLU results to compare against other models
- Weights you can fine-tune on institutional documentation and terminology
- A model preference-tuned to answer clinical questions rather than refuse them
- A technical report covering training data, hyperparameters and evaluation
Med42 is a credible, free clinical base model from a large Gulf health group, published on Hugging Face rather than behind a sales call, and a real option for a health system that wants to run a medical model on its own hardware.
Discount the scores: the USMLE figures are M42's own, and exam performance says little about real clinical notes. It is built on Meta's Llama 3 and its licence, has not been updated since 2024, has no hosted product, and M42 says it "is not ready for real clinical use." Treat it as something to fine-tune and validate.
What we found
Med42 is a family of clinical language models from M42, the Abu Dhabi health group formed when G42 Healthcare and Mubadala Health came together. The first release, in October 2023, was a 70B fine-tune of Meta's Llama 2, trained on Cerebras hardware and G42's Condor Galaxy supercomputer in collaboration with Core42. M42 reported a 72 per cent zero-shot score on USMLE sample questions and released the weights on Hugging Face for non-commercial and research use.
Med42-v2 followed in 2024. M42 announced a 70B Llama 3 model in May and later published both 8B and 70B weights under the Llama 3 Community License, with a technical report in August. The press release puts the 70B at 85.1 per cent zero-shot on USMLE questions and 87.3 per cent with specialised prompting, "on par" with GPT-4 and Med-Gemini; the model card's own table gives 83.8 per cent zero-shot. Both are M42's own measurements on exam questions, which say little about how a model handles real clinical notes. An earlier version of this page credited v2 to Cerebras training, which described v1.
Nothing newer has appeared. The Med42 weights on Hugging Face were last updated in August 2024, and M42's recent public models cover other areas, including chest X-rays and genomics. We found no hosted Med42 service and no regulatory clearance, and the model card states plainly that the suite "is not ready for real clinical use."
What Med42 offers is a capable, downloadable clinical base model that a hospital can run on its own hardware, keeping patient data in-house. The weights depend on Meta's base model and licence, so its independence is institutional rather than technical. The 70B version needs substantial GPU capacity, while the 8B is far lighter and scores lower. Treat it as a starting point to fine-tune and validate, not something to hand to clinicians as shipped.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Med42's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Med42 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 Med42 directly →
Spotted something out of date? Suggest an update →
Alternatives to Med42
More in Healthcare
