Google DeepMind · 2026-04-30 · major
DeepMind AI Co-Clinician — Talker/Planner Dual-Agent Hits Zero Critical Errors on 97 of 98 Primary Care Queries
DeepMind unveiled AI co-clinician, a Gemini- and Project Astra-based system for triadic care. A Planner module supervises a Talker agent in real-time multimodal patient conversations. In blind physician evals it surpassed primary care doctors on 68 of 140 consultation skills.
DeepMind's clinical AI uses a Planner agent to keep a Talker agent inside safe medical boundaries during live audio/video patient consultations.
Key specs
| Critical error free queries | 97 / 98 |
|---|---|
| Consultation skills surpassing pcps | 68 / 140 |
| Partner countries | 6 (US, India, Australia, NZ, Singapore, UAE) |
What is it?
AI co-clinician is a research system from Google DeepMind built on Gemini and Project Astra capabilities. It is designed for what the team calls triadic care: an AI agent assists a patient under the clinical authority of a supervising physician. The system handles live audio and video, performs evidence synthesis with citation-checking, and is currently in phased research collaborations with Harvard Medical School, Stanford Medicine, and academic centers in India, Australia, New Zealand, Singapore, and the UAE. It is not yet available for clinical use.
How does it work?
AI co-clinician uses a dual-agent architecture: a Talker model interacts with the patient while a Planner module continuously monitors the conversation to ensure responses stay within safe clinical boundaries. The system pulls verified citations from a clinical knowledge corpus and supports real-time multimodal input (live audio and video) for telemedical simulations. DeepMind released a technical report titled 'Towards Conversational Medical AI with Eyes, Ears and a Voice' alongside the announcement.
Why does it matter?
The benchmark numbers are unusual for a clinical AI: zero critical errors in 97 of 98 realistic primary care queries, and head-to-head wins over primary care physicians on 68 of 140 consultation skills in blind evaluation. The team explicitly frames this as augmentation, not replacement: 'Medicine has always been a team sport, and AI agents can bring more teammates onto the field: extending clinicians' reach while ensuring they retain judgment and control.' Pairing a supervisor agent with a conversational agent is a pattern other safety-sensitive deployments may adopt.
Who is it for?
Clinical AI researchers, health-system CTOs evaluating AI deployment patterns
Try it
Read the technical report at gstatic.com/vesper/ai_coclinician_technical_report.pdf