A foundation model for human biology.

Predict what will happen to any patient at any time. Standard Model Biomedicine builds a foundation model for human biology — a medical AI trained across genomic, imaging, and longitudinal patient records to forecast clinical outcomes.

What the Standard Model does

The Standard Model turns everything generated for one patient — assessment, imaging, pathology, genomics, interventions — into a single patient-level representation, then forecasts what will happen under any intervention.

Adapt it to your endpoint

Partners adapt the model to any clinical endpoint — survival, treatment response, toxicity, patient stratification — with a lightweight task head trained on their own labels. No foundation-model retraining is required; see use it on your data.

Open-weight models

The model collection

Patient encoders and modality-specific encoders for EHR, CT, genomics, and proteomics are published as open weights on Hugging Face, with model cards and clinical evaluations. Browse the model hub for the full collection, including the open JEPA model for longitudinal EHR.

Run them on your machine

Download the weights and produce patient-level embeddings you can feed into your own downstream models. The install guide and the end-to-end example walk through a complete run — from raw data to a linear-probe evaluation — in about a minute on a single GPU.

Research and team

Our published research spans clinical foundation models, medical imaging, genomics, and federated learning for drug development — the methods, models, and clinical evaluations underwriting the platform. The team comes from BMS, Genentech, AbbVie, Sanofi, Stanford, MIT, Yale, UCSD, MD Anderson, and Mount Sinai.

Explore this site

Company and contact

Standard Model Biomedicine is based in San Francisco, California. Reach us at info@standardmodel.bio for partnerships, model access, research collaborations, or press.