[I/O 2026] Google introduces Gemini for Science, a series of agentic tools for scientific research

Google launches Gemini for Science on Google Labs, an agentic suite featuring Co-Scientist and NotebookLM to automate hypotheses and literature analysis.

Gemini for Science brings together several tools designed to accelerate the core stages of the scientific method. Three experimental prototypes are progressively launching on Google Labs. Hypothesis Generation, built on Co-Scientist, simulates an idea tournament among multiple agents to generate, debate, and evaluate hypotheses, with each assertion supported by verifiable citations. Computational Discovery, based on AlphaEvolve and the empirical research assistant ERA, simultaneously generates and evaluates thousands of code variants to test modeling approaches in fields such as solar forecasting or epidemiology. Literature Insights, powered by NotebookLM, structures scientific literature into comparable tables and helps identify gaps and unexplored avenues.

In parallel, Google is making Science Skills available, a suite that aggregates over thirty life sciences databases and tools, including UniProt, the AlphaFold database, the AlphaGenome API, and InterPro. Integrated with agentic platforms like Google Antigravity, this system allows for structural or genomic bioinformatics analyses to be conducted in minutes rather than hours. During internal tests, it helped shed light on possible mechanisms of a rare genetic disease linked to mutations in the AK2 gene.

The deployment is accompanied by a professional dimension via Google Cloud, already in preview with partners such as BASF, Klarna, Daiichi Sankyo, Bayer Crop Science, and US national laboratories. Research work associated with ERA and Co-Scientist is published in Nature. Over one hundred institutions, including Stanford, Imperial College London, and the Crick Institute, are participating in the validation, alongside a community of testers ranging from PhD students to Nobel laureates.