BioMedLM

BioMedLM is a 2.7-billion-parameter GPT-style language model from Stanford's Center for Research on Foundation Models, built with MosaicML and trained exclusively on PubMed abstracts and full-text articles. Previously known as PubMedGPT, it was an early demonstration that smaller domain-specific models could rival much larger general models on specialized tasks, reaching 50.3 percent on MedQA biomedical question answering at release.

The model uses a standard GPT-2 architecture and was released openly on the Hugging Face Hub to advance biomedical NLP and responsible practices for domain-specific training. Stanford emphasizes it is for research only, not clinical use: generation capabilities are exploratory, and questions of reliability and truthfulness in medical applications remain open. It has become a reference point for transparent, privacy-preserving biomedical models that run without sending data to third-party APIs.