Last updated: 2026-07-27
Google's Gemini for Science pairs hypothesis generation, agentic experiments, and NotebookLM review; ERA beat CDC's COVID forecast in Nature-reviewed tests.
Gemini for Science is Google's suite of agentic research tools launched at I/O 2026, built on Co-Scientist, AlphaEvolve, ERA, and NotebookLM. Its Empirical Research Assistance engine generated 14 COVID-19 forecasting models that beat the CDC's CovidHub Ensemble, validated in two peer-reviewed Nature papers.
Gemini for Science is a suite of AI research tools that Google introduced at I/O 2026, built by Google DeepMind and Google Research to speed up the slowest parts of the scientific method for working scientists. It packages four components, Hypothesis Generation, Computational Discovery, Literature Insights, and Science Skills, so a single researcher can move from a research question to a testable model in far fewer manual steps than reading, coding, and testing by hand allows. Hypothesis Generation runs on Co-Scientist and stages a multi-agent idea tournament: separate AI agents propose competing hypotheses, debate their novelty and feasibility, and score them against the existing literature before a researcher ever writes code. Computational Discovery then hands the winning idea to AlphaEvolve and Empirical Research Assistance (ERA), which generates and runs thousands of code variations in parallel to test many modeling approaches at once. Two papers published in Nature alongside the launch reported that ERA produced 14 COVID-19 hospitalization forecasting models that beat the CDC's own CovidHub Ensemble, plus new single-cell analysis methods validated against public bioinformatics leaderboards. Literature Insights, built on NotebookLM, turns a pile of papers into a searchable table and a conversational interface, and can output the summary as a written report, a slide deck, an infographic, or an audio or video briefing. Science Skills is aimed squarely at life sciences teams, wiring in dozens of databases and tools such as UniProt, the AlphaFold Database, AlphaGenome, and InterPro so a computational biologist can query protein structure and genomic data without switching tools. The suite targets academic labs, life sciences R&D teams, and epidemiology or climate researchers who need to test many modeling approaches quickly rather than one at a time. Access to Gemini for Science works through a gradual, request-based rollout on Google Labs, without a published consumer price during the current preview. Enterprises can reach the same Co-Scientist and AlphaEvolve agents in production via Google Cloud, specifically its Gemini Enterprise Agent Platform product, on custom contract terms. The tools run in-browser; there is no dedicated desktop or mobile app. Both Nature papers underpinning ERA were peer-reviewed ahead of the product announcement, a step Google took specifically to back the platform's benchmark claims with independent review rather than marketing copy alone.
Free during the Google Labs preview (request-only access, no self-serve paid plan). Google Cloud enterprise deployment of Co-Scientist and AlphaEvolve through the Gemini Enterprise Agent Platform is billed on custom, unpublished contract terms.
Google has not published a retail price for Gemini for Science. The Google Labs experiments are free to request during the gradual rollout, while the underlying Co-Scientist and AlphaEvolve agents are also offered to organizations through Google Cloud's Gemini Enterprise Agent Platform on custom contract pricing that Google does not publish as a plan.
Yes, for individual researchers: entry to the Labs experiments is currently at no cost, gated behind a request form rather than a self-serve signup, and rolling out gradually rather than to everyone at once. There is no published usage cap because the product has not left the request-access stage.
Consensus and NotebookLM both cover the literature-review half of the workflow if you don't need hypothesis generation or experiment automation. ChatGPT and Perplexity work as general-purpose research assistants but lack a dedicated agentic experiment runner like AlphaEvolve and ERA or the peer-reviewed benchmark backing Google published in Nature.
Consensus is built specifically for fast literature search and answer-finding across papers, with instant self-serve access. Gemini for Science covers that same literature step through its NotebookLM-based Literature Insights tool, then goes further into hypothesis generation and parallel computational experiments, at the cost of a request-only rollout instead of instant signup.
Visit labs.google/science and submit the access request form; Google reviews requests as part of the gradual rollout rather than granting instant access. Enterprises that want Co-Scientist or AlphaEvolve as production agents should instead contact Google Cloud about the Gemini Enterprise Agent Platform.
Gemini for Science pairs a hypothesis-tournament agent, an AlphaEvolve experiment runner, and NotebookLM review, backed by two Nature papers at I/O 2026.
Google ยท Free tier available