MEDITRON

MEDITRON is the open medical LLM family from EPFL's NLP/MLO labs and Yale School of Medicine, built with humanitarian medicine in mind (ICRC clinical guidelines are in the training mix). The 7B and 70B models continue pretraining Llama-2 on a curated medical corpus — PubMed literature plus international clinical practice guidelines — and MEDITRON-70B outperformed GPT-3.5 and Med-PaLM on medical benchmarks, landing within 5% of GPT-4. Weights, corpus-curation code and the newer MeditronFO fully-open training pipeline are all public.

Important caveats: it is a research model, not a certified medical device — no LLM should drive clinical decisions unsupervised. The base 70B inherits Llama 2's community license terms, and as a 2023-era model its general capabilities lag modern instruct models. But for researchers who need an auditable, openly-trained medical foundation — especially for low-resource and humanitarian settings — it remains the reference open medical LLM.