The Standard Model™ V-2

Powered to predict complex patient progression and the most effective signatures, biomarkers and trial designs.

JEPA World Model embedding, encoder and decoders.

A completely novel architecture allows the Standard Model to be rapidly deployed, trained, fine-tuned and powered with external foundation models and proprietary data.

Molecular
Germline & somatic DNA, RNA expression, histopathology images
Patient-level
CT / MRI, electronic health records

The Standard Model™ V-2 is lightweight and easy to deploy in your AI environment.

A compact test set beat trillion‑parameter frontier AI in independent benchmarking.

  • 3 billion Tuned Parameters
  • 2 sec p50 Inference
  • 8 sec Average patient inference time
  • ~50k Patients used for tuning
  • <$0.01 Per Patient
  • 0 Network calls

Standard Model beats frontiers for patient inferences.

Model Cost per prediction (USD) Median (p50) inference time
Standard Model$0.0092 seconds
Gemini-3.7-flash$0.32108 seconds
Claude-Opus-5 (max reasoning)$1.50166 seconds
GPT-5.6-sol (max reasoning)$1.53108 seconds

Running the benchmark’s 2,254 predictions cost about $20 on Standard Model versus between $720–$3,450 on the frontier models.

The Standard Model is run in a completely secure container with no opportunity for bleed‑out.

See the Protege independent benchmarking data Get the Standard Model™ V-2
Assessment Imaging Pathology Genomics Intervention THE STANDARD MODEL MULTIMODALPATIENT ENCODER Observed patient history NOW RESPONSEResponder probability0.74HORIZON • 1 CYCLE STRATIFICATIONSubgroup assignmentHORIZON • 6 MONTHS YOUR ENDPOINTReadmission • DLT • CustomHORIZON • ANY TOXICITYGrade ≥3 AE risk0.22HORIZON • 3 MONTHS SURVIVALOS/PFS curveHORIZON • 12 MONTHS SHARED DECODERTrainable heads / reads the rollout atany horizon Observed Patient State Interventions Predicted Patient State Endpoint read-outthe shared decoder emits it at that horizon

Model clinical endpoints using minimal inputs.

Identify patients likely to qualify for future trials
External / synthetic control arms for single-arm registrational trials; models disease projection so sponsors don’t wait years for an endpoint the model can estimate now.
Time to next therapy
The pragmatic, payer-legible outcome sponsors lean on when adverse event data has not been established, providing a direct competitor to manual RWE chart-review studies.
Disease progression
Automated response classification — an independent or supplemental read alongside (or instead of) a central imaging core lab.
Combination & comparators
Digital twins to model in silico response against comparators in real time, enabling modeling of comorbidities and prior treatments.
Biomarker prediction & selection
Simple and multimodal signatures and selection of the best biomarkers including cut-offs and other meaningful parameters.

Model access

  • Simple licensing with success‑driven opt‑in and opt‑out
  • Deployed in your AI environment
  • Supported by Standard Model’s pharma‑native AI Applications Scientists
Get the Standard Model™ V-2