Skip to main content

Services

Clinical Foundation Models

Domain-specific AI models from 7B to 400B+ parameters. Pre-trained on clinical data, fine-tuned for your use case.

MedBase Model Family

Four tiers scaled to your use case, infrastructure, and budget. All models share the same clinical pre-training base and validation suite.

Lightweight

MedBase-7B

Clinical named entity recognition, entity extraction, and structured data parsing. Runs on a single consumer GPU. Ideal for annotation pipelines and high-throughput extraction tasks.

  • Single GPU
  • Clinical NER
  • Entity Extraction
  • Fast inference
Documentation

MedBase-13B

Clinical documentation, visit summarization, and decision support. Balances capability and deployment cost. Suitable for EHR-integrated workflows.

  • 1–2 GPUs
  • Summarization
  • Documentation AI
  • Decision support
Reasoning

MedBase-70B

Complex clinical reasoning, differential diagnosis support, and research synthesis. Handles nuanced multi-step reasoning across long clinical contexts.

  • Multi-GPU
  • Differential Dx
  • Research synthesis
  • Long context
Enterprise

MedBase-400B+

Maximum capability for enterprise clinical AI deployment. Full multi-specialty coverage, highest benchmark scores, and custom fine-tuning on request.

  • GPU cluster
  • Max capability
  • All specialties
  • Custom fine-tuning

What Makes These Different

Clinical pre-training — not scraped web text

MedBase models are pre-trained on synthetic clinical data from our hybrid pipeline. The training corpus reflects real clinical distributions: patient trajectories, documentation patterns, diagnostic reasoning, and treatment sequences — not blog posts and Wikipedia.

Validated against clinical benchmarks

Every MedBase tier is evaluated against MedQA, PubMedQA, Clinical NER, and BioASQ before release. We publish target benchmark numbers so you can compare against general-purpose models directly. No vague capability claims.

On-premise by design

MedBase models are built for on-premise deployment. No cloud dependency, no outbound API calls, no per-query licensing. You run the inference. Your data never leaves your environment.

Custom Fine-Tuning

Need a model tuned to your specialty, formulary, or patient population? We fine-tune any MedBase model on your de-identified data. The result is a private model you own — calibrated to how your organization documents, diagnoses, and treats.

Fine-tuning engagements follow the same data sovereignty principles as our hospital AI service: your data trains the model, the model ships to you, we retain nothing.

Licensing Options

Academic & Non-Commercial

Research

For universities, research institutions, and non-commercial projects. Includes model weights, documentation, and standard benchmarks.

  • Model weights (any MedBase tier)
  • Research use only
  • Standard documentation
  • Community support

Production Deployment

Commercial

Production deployment within your organization. Unlimited queries, on-premise or private cloud, standard SLA.

  • Single-organization deployment
  • On-premise or private cloud
  • Standard SLA
  • Email support

Unlimited + Custom Fine-Tuning

Enterprise

Unlimited deployment across your organization or products, priority support, and custom fine-tuning on your data included.

  • Unlimited deployment
  • Custom fine-tuning included
  • Priority support SLA
  • Dedicated account team

Target Benchmarks

Target performance figures. MedBase-70B and 400B+ tiers. General-purpose baseline reflects current frontier models on clinical evaluations.

BenchmarkGeneral-PurposeMedBase Target
MedQA (USMLE-style)62–68%84%+
PubMedQA60–72%88%+
Clinical NER (F1)65–75%91%+
BioASQ58–70%83%+

* Target figures. Actual results vary by model tier and fine-tuning. Independent evaluations published with each model release.

Need a clinical model for your stack?

Discuss licensing, fine-tuning, or custom deployment with our team.

Get in Touch