Gemini for Science: AI experiments and tools for a new era of discovery
Google DeepMind ● Covered by 2 sources
Google just launched Gemini for Science, a bundle of AI tools meant to speed up research from idea to lab result. It's already helping partners like BASF and Bayer, and it just found a genetic disease clue in minutes.
Based on reporting by Google DeepMind — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind has a new pitch for science: stop building narrow AI models for single problems and instead hand researchers general-purpose agents that can chew through the whole scientific method. That's the idea behind Gemini for Science, a set of tools rolling out today through Google Labs and Google Antigravity, and it's a bigger swing than the usual incremental research-assistant update.
Three experimental prototypes anchor the release. Hypothesis Generation, built on Co-Scientist, runs what Google calls an idea tournament — multiple AI agents propose, argue over, and rank research hypotheses, with citations attached so scientists can check the sourcing. Computational Discovery pairs AlphaEvolve with a system called ERA to generate and score thousands of code variations in parallel, aimed at fields like epidemiology or solar forecasting where testing every possible model by hand would eat months. And Literature Insights, built on NotebookLM, turns piles of papers into searchable comparison tables, plus slide decks and audio summaries, so researchers can spot gaps in a field faster than by reading everything themselves.
The more concrete proof point is Science Skills, a bundle inside Google Antigravity that plugs into more than 30 life-science databases, including UniProt, the AlphaFold Database, and the AlphaGenome API. Google says its own researchers used it to run a structural bioinformatics analysis that normally takes hours, finishing in minutes, and in the process turned up a new lead on how mutations in the AK2 gene might cause a rare genetic disease. That's a small result on its own, but it's the kind of specific, checkable claim that separates this from typical AI-for-science marketing.
Google isn't just giving this away for free exploration, either. BASF is already using AlphaEvolve to optimize supply chains, Klarna is applying it to machine learning models, and Co-Scientist is in private preview at Daiichi Sankyo, Bayer Crop Science, and US National Labs under the Department of Energy's Genesis Mission. Two papers on ERA and Co-Scientist are publishing in Nature today, which suggests Google wants peer-reviewed credibility attached to this launch, not just a blog post and a waitlist.
There's also a validation layer worth noting: more than 100 institutions, from Stanford to Imperial College London to the Crick Institute, are testing these systems, alongside a
My take — AI-written commentary, not fact-checked reporting
I'll believe the AK2 gene story once it's independently replicated, because Google has a long habit of pairing genuinely useful tools with anecdotes engineered for headlines. That said, pairing this launch with actual Nature papers and named enterprise customers is a step up from the usual vague 'AI accelerates science' hand-waving — I'd rather see ten boring, verifiable wins like a faster literature review than one flashy unproven cure. Watch whether independent labs outside Google's trusted-tester program get equal access, because that's the real test of whether this is open science infrastructure or just Google's latest walled garden with a lab coat on.
Read more about this at: Google DeepMind