Google DeepMind · 2026-05-19 · major
Google DeepMind Co-Scientist Lands in Nature — Multi-Agent Gemini System Generates, Debates, and Evolves Scientific Hypotheses With 100+ Research Partners
Co-Scientist orchestrates seven Gemini-powered agents through a generate / debate / evolve loop. Nature publishes the paper alongside a Hypothesis Generation tool in Google Labs and an enterprise version on Google Cloud.
Seven Gemini agents argue, rank, and refine scientific hypotheses — now a Nature paper and a Google Labs tool.
Key specs
| Specialized agents | 7 |
|---|---|
| Partner institutions | 100+ |
| Published in | Nature |
What is it?
Co-Scientist is a multi-agent system built on Gemini that proposes research hypotheses, critiques them through simulated peer review, and evolves the survivors. Google launched it in three forms: a published Nature paper, a Hypothesis Generation tool at labs.google/science, and an enterprise offering through Google Cloud.
How does it work?
Generation and Proximity agents propose and map ideas grounded in literature. Reflection and Ranking agents run pairwise debate tournaments to score them. Evolution and Meta-review agents merge winners and synthesize the final survey. A Supervisor agent coordinates parallel exploration. Researchers iterate by feeding back domain critiques.
Why does it matter?
Tested across 100+ institutions, Co-Scientist has produced hypotheses that match unpublished lab discoveries and proposed antimicrobial-resistance leads, ALS approaches, and liver-fibrosis treatments. Daiichi Sankyo, Bayer Crop Science, and US National Labs are listed enterprise partners. It is the most ambitious peer-reviewed test of agent science to date.
Who is it for?
Research scientists, drug discovery teams, computational biology groups
Try it
https://labs.google/science