Med42

Med42 is an open-access clinical large language model from M42, the Abu Dhabi health-tech group (G42). The 70B model is fine-tuned from Llama-2 on a curated medical corpus, trained with Cerebras in collaboration with Core42 and evaluated with MBZUAI. M42 reported 72% zero-shot on USMLE sample questions, ahead of ChatGPT 3.5 at the time (vendor-reported).

A Med42-v2 suite on Llama 3 (8B and 70B) followed with two-stage clinical fine-tuning and preference alignment, reporting up to 94.5% on some medical benchmarks (vendor-reported). Treat it as a research artifact for clinical NLP, not a certified clinical tool; medical deployment needs proper validation and oversight.