Google DeepMind published research on April 30 detailing an AI co-clinician built on Gemini and Project Astra, evaluated via live telemedical simulations with real physician "patient-actors." In blind evaluations of 98 primary care queries, the system beat GPT-5.4-thinking-with-search 63–30 and achieved zero critical errors in 97 of 98 cases; it also performed at or above primary care physician level in 68 of 140 assessed consultation skills. The system uses a dual-agent "Planner/Talker" design in which a supervisor agent continuously checks whether the Talker stays within clinical boundaries, but experienced physicians still outperformed it on identifying red flags and guiding critical physical exams. The technical report, produced in collaboration with Harvard Medical School and Stanford Medicine, is titled "Towards Conversational Medical AI with Eyes, Ears and a Voice."