About Standard Model Biomedicine

Standard Model Biomedicine builds the world model of patients and in silico trials, for drug developers and clinicians.

What is the Standard Model?

No single company can build all the data and AI that biomedicine needs on its own. So we build alongside the leading LLMs, frontier models and patient atlases, then add our own proprietary data and model architecture. The result is one joint embedding of the patient: a single architecture for encoding and decoding patient data.

The Standard Model’s outputs are digital representations of patients that predict how each patient’s biology will change in response to treatment. That lets teams model many aspects of a clinical trial’s design years before it reaches its endpoints.

In practice, the Standard Model can:

• Model clinical trial design, allowing every potential trial input to be modeled against the others in a phase zero setting
• Provide in silico digital twins that predict patients’ journeys across indication, disease and time
• Project treatment outcomes, disease progression times on treatment & overall survival
• Provide prognostic suggestions of the next best treatment

SMB has built one of the world’s first Joint Embedding Predictive Architectures (JEPA) of the patient. First, we merged the top-performing purpose-built models into a single patient representation. Then we trained the multimodal Standard Model on our proprietary data across 8 modalities, including DNA, RNA, CT images, whole slide images and other biomarkers. Finally, the model encodes and decodes each patient, drawing on longitudinal electronic health records from millions of patients.

How the Standard Model works

Put simply, the Standard Model turns everything generated for one patient (assessment, imaging, pathology, genomics and interventions) into a single patient-level representation, then predicts what will happen under any intervention. Users then adapt it to their own clinical endpoints, such as survival, response or toxicity, with a lightweight task head trained on their own labels. There’s no need to retrain the underlying model.

The Standard Models

The Standard Model™ V-2 (available through early access) is a JEPA model that combines multimodal encoders into one patient model, built with leading academic and industry partners. It supports indication screening, dose optimization, toxicity prediction, regulatory preparation, outcome prediction and zero-shot inference. V-2 is accessed through a lightweight user agreement and deployed directly in your AWS or other secure environment. Unlike other licensed models, it can be used off the shelf or fine-tuned.

TrialFront™ is our in silico trial designer, built on the Standard Model. It turns a study brief into 10–20 candidate trial designs, simulates each one across thousands of virtual trials, and ranks them by probability of success, duration and cost. See how TrialFront works.