On May 19, 2026, alongside Google I/O 2026, Google unveiled "Gemini for Science," a collection of AI tools aimed at researchers[1]. The lineup pairs three Google Labs prototypes — a Co-Scientist–powered hypothesis generator, an AlphaEvolve-based computational research engine, and a NotebookLM-derived literature analysis tool — with "Science Skills," a bundle of more than 30 life-science databases delivered through Google's agent development platform, Antigravity. Together, the pieces push AI a step closer to being a partner that actually runs work on behalf of scientists[1][2].
Three research prototypes rolling out on Google Labs
At the heart of Gemini for Science are three experimental tools that Google Labs will open up gradually starting today[1].
The first is Hypothesis Generation, built on top of DeepMind's collaborative AI agent "Co-Scientist." Researchers first frame a research question, after which a multi-agent "idea tournament" generates, debates and scores hypotheses end to end[1][3]. Each claim it returns is deeply verified and accompanied by clickable citations, a design choice meant to keep hallucinations from slipping into the research workflow[1].
The second is Computational Discovery, an agentic research engine that combines DeepMind's AlphaEvolve with ERA (Empirical Research Assistance), whose validation paper was published today in Nature[1][4]. By generating and scoring thousands of code variants in parallel, it tackles compute-heavy fields like solar forecasting and epidemiology, compressing months of manual modeling into a much shorter cycle[1].
The third is Literature Insights, built on NotebookLM. It searches scientific literature, then organizes the results into tables with custom, searchable attributes so papers can be compared side by side[1]. From there, researchers can drill into nuances through chat and produce high-fidelity artifacts — reports, slide decks, infographics, even audio and video overviews — which sets it apart from traditional survey workflows[1].
Across all three tools, Google says access will roll out in stages starting today, with interested researchers invited to register at labs.google/science[1].
"Science Skills" on Antigravity bundles 30+ life-science databases
The other pillar is Science Skills[1]. It packages more than 30 major life-science databases and tools — including UniProt, AlphaFold Database, AlphaGenome API and InterPro — into a skill pack that runs on Google's agent development platform Antigravity. According to Google, manual workflows such as structural bioinformatics and genomic analyses that typically take hours can now be completed in a matter of minutes[1].
In internal testing, Google's team used Science Skills to analyze a rare genetic disease caused by mutations in the AK2 gene, surfacing new hypotheses about its underlying mechanism[1]. Researchers can try Science Skills directly inside Antigravity at antigravity.google/use-cases/science[1].
From experiments to the enterprise and academic frontline
Gemini for Science is not limited to experimental tools. Through Google Cloud, it is also being rolled out for enterprises: BASF is using AlphaEvolve to optimize its supply chain, Klarna is applying it to strengthen its machine learning models, and Daiichi Sankyo, Bayer Crop Science and the U.S. National Laboratories participating in the U.S. Department of Energy's Genesis Mission are using Co-Scientist to accelerate their research[1].
Ties to academia have also deepened. Google says it is collaborating with more than 100 institutions, including Stanford University on liver fibrosis, Imperial College London on antimicrobial resistance and a multi-year program with the Francis Crick Institute, putting AI to work on tough life-science problems[1]. A trusted-tester community spanning PhD students to Nobel laureates is stress-testing the systems against complex real-world challenges — an important safeguard for the credibility of AI-generated findings[1].
Today's news also coincides with the publication of the ERA and Co-Scientist validation papers in Nature, alongside agentic peer-review experiments such as Paper Assistant Tool and ScholarPeer launching with ICML, STOC and NeurIPS — a sign that AI is stepping into every stage of the research loop[1]. With AlphaFold already used by over 3 million researchers and AlphaGenome helping identify disease drivers, Gemini for Science looks less like a one-off campaign and more like a convergence point for Google's accumulated science-oriented AI work[1].
Conclusion
Gemini for Science, unveiled at I/O 2026, is a two-tier offering — three Labs prototypes plus Antigravity's Science Skills — that pushes agentic AI directly into the research process[1]. Backed by enterprise and academic deployments at BASF, Daiichi Sankyo and the U.S. National Labs, and reinforced by the Nature validation papers, it underscores a broader shift: AI is moving from "helping with research" to actually "driving research" forward[1][4].
Source [1]: https://blog.google/innovation-and-ai/technology/research/gemini-for-science-io-2026/
Source [2]: https://ai.google/gemini-for-science/
Source [3]: https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/
