source: google research: a new era of innovation: google research at i/o 2026
level: technical
google research unveiled gemini for science, a suite of experimental tools to accelerate scientific discovery. it includes computational discovery, built with empirical research assistance (era) and alphaevolve, which generates and scores thousands of code variations to test hypotheses rapidly. another tool, hypothesis generation, uses the co-scientist multi-agent system to run idea tournaments that generate, debate, and evaluate hypotheses with clickable citations. literature insights, built with notebooklm, helps synthesize findings across papers. these tools aim to help researchers tackle problems from neuroscience to cosmology, with era already predicting hospital admissions and seasonal runoff.
in health, google is rolling out the google health app to fitbit users, offering personalized coaching based on a personal health llm. new research includes symptom ai, which in a study of over 13,000 participants produced differential diagnoses preferred by clinicians about twice as often as other clinicians. a plan for care pilot helped 15% more users feel prepared for doctor visits. in clinical settings, the amie system now handles multimodal data like lab results and medical images, with real-world testing at beth israel deaconess medical center. medgemma, an open-weight model for medical text and imaging, has surpassed 5 million downloads.
other highlights include the coralboard from synaptics, featuring google's coral npu for energy-efficient edge ai, demonstrated with live jellyfish detection at the monterey bay aquarium. for crisis resilience, weathernext accurately predicted hurricane melissa's rapid intensification five days ahead, aiding jamaica's early warnings. a new method called groundsource used gemini to turn 20 years of news reports into a dataset of 2.6 million records, enabling urban flash flood forecasts now available on flood hub. google also noted ongoing work on llm factuality and core model improvements with deepmind.
why it matters: these tools and models can help scientists, doctors, and developers solve real problems faster, from drug discovery to disaster response, while making ai more reliable and accessible.
source: google research: a new era of innovation: google research at i/o 2026