Google / DeepMind · Healthcare

Med-PaLM 2

Google's medical large language model that achieves expert-level performance on medical licensing exam questions and clinical reasoning tasks.

Overview

Med-PaLM 2 builds on Google's PaLM 2 foundation model with specialized medical instruction tuning and alignment. It was the first AI system to reach expert-level performance on the U.S. Medical Licensing Examination (USMLE), scoring above 85%. The model incorporates ensemble refinement and chain-of-thought prompting strategies tailored for medical reasoning, making it a leading system for clinical question answering and medical consultation support.

Base Model

PaLM 2

USMLE Score

86.5% (expert level)

Training Approach

Instruction tuning + ensemble refinement

Availability

Limited access via Google Cloud

Modality

Text (multimodal variant available)

Capabilities

Medical question answering at expert level

Clinical reasoning and differential diagnosis

Medical exam preparation and assessment

Patient-friendly medical explanation generation

Multi-step medical reasoning with chain-of-thought

Medical literature synthesis

Use Cases

Supporting clinical decision-making with evidence-based answers

Generating patient education materials in plain language

Assisting medical students with exam preparation

Providing second opinions on complex diagnostic cases

Pros

  • +Expert-level performance on medical licensing exams
  • +Strong clinical reasoning with explainable chain-of-thought
  • +Backed by Google/DeepMind's research infrastructure
  • +Aligned to reduce harmful medical advice generation

Cons

  • -Not publicly available; restricted access through Google Cloud
  • -Closed-source with no ability to self-host
  • -High cost for enterprise deployment
  • -Not FDA-approved for direct clinical decision-making

Pricing

Available through Google Cloud with enterprise agreements. Pricing is custom and requires direct engagement with Google's healthcare AI team.

Related Models