Med42
Abu Dhabi's open-access clinical LLM, scoring in GPT-4 territory on USMLE questions
Overview
Med42 is an open-access clinical large language model from M42, the Abu Dhabi health group. First released in October 2023, the Med42-v2 generation announced in May 2024 is a Llama-3 based family in 8B and 70B sizes, trained with Cerebras in collaboration with Core42 and published on Hugging Face.
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 These figures reflect RECATOOLS research as of 4 Sep 2026 and may be out of date or incomplete.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 Med42
- Download open-access clinical model weights in 8B and 70B sizes from Hugging Face
- Run a clinical assistant entirely inside your own infrastructure with no third-party inference calls
- Compare published USMLE-style benchmark results against the proprietary models you are considering
- Fine-tune the released weights on your own institutional documentation and terminology
- Draft and summarise clinical documentation with a model trained specifically on medical text
- Evaluate a clinical model built outside the US and Chinese vendor duopoly, under Gulf governance
Med42 matters because it is open, credible and not American. M42 published Llama-3-based clinical weights at 8B and 70B with USMLE results the company puts alongside GPT-4, and put them on Hugging Face rather than behind a sales call — a real option for any health system that cannot ship patient data to a US API.
Discount the exam scores appropriately; USMLE performance has a poor record of predicting behaviour on real clinical notes. There is no supported hosted product, no clearance, and thinner documentation than a commercial vendor would provide. Treat it as a strong base model to build and validate on, not as something you deploy to clinicians as shipped.
What people say
Med42 is the clearest evidence that frontier-adjacent medical models no longer have to come from Silicon Valley. M42, the Abu Dhabi health group, put the first version out in October 2023 and followed it in May 2024 with Med42-v2, a Llama-3 family in 8B and 70B sizes trained with Cerebras alongside Core42. The weights went to Hugging Face rather than behind a sales process, which is what makes it worth evaluating rather than merely noting.
The numbers M42 published are strong: 85.1% zero-shot on USMLE sample questions, rising to 87.3% with specialised prompting, which the company positions as comparable to GPT-4 and Med-Gemini. Treat those the way you would treat any vendor-published benchmark — they are exam questions, and exam performance has repeatedly failed to predict how a model behaves on messy clinical notes from a real health system. What the scores do establish is that the gap between open medical weights and closed ones is narrower than most procurement conversations assume.
The gaps are institutional rather than technical. There is no hosted API sold as a product, documentation is thinner than a commercial vendor's, and the model carries no regulatory clearance for clinical decision-making — it is a foundation to build on, not a device. The 70B model needs serious GPU capacity, and the 8B version trades accuracy for the ability to run somewhere modest.
For a hospital group that cannot send patient data to a US-hosted API, or for a team that wants a credible clinical base model to fine-tune, Med42 is one of a very small number of genuine options. It also carries a governance story that matters to some buyers: the model, the compute and the institution all sit outside the two blocs that dominate the rest of this list.
Summary of public user & expert reviews, compiled by RECATOOLS.
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
