Gemini for Science: Google’s New AI Toolkit for Accelerating Scientific Discovery

Together, these components allow AI systems to retrieve scientific knowledge, analyze research, propose new hypotheses, and run computational experiments with minimal manual setup.

Core Tools in the Gemini for Science Suite

Co‑Scientist

Co‑Scientist acts as an AI research collaborator designed to help scientists generate new hypotheses and research proposals. Built as a multi‑agent system using Gemini models, it mirrors elements of the scientific reasoning process and assists with literature analysis and research planning.

Researchers can specify a research goal—such as understanding a disease mechanism—and the system proposes potential hypotheses along with summaries of relevant literature and experimental approaches.

AlphaEvolve

AlphaEvolve is a Gemini‑powered evolutionary algorithm agent designed to discover optimized algorithms for complex scientific and engineering problems.

The system iteratively tests and refines algorithmic solutions, enabling it to tackle problems that traditionally require large amounts of manual experimentation. According to Google, AlphaEvolve has already been applied to areas such as:

• Improving DNA sequencing error correction
• Increasing accuracy of disaster predictions
• Simulating techniques to help stabilize power grids

These applications demonstrate how algorithm‑search agents can contribute to scientific and infrastructure research.

ERA (Empirical Research Assistance)

Empirical Research Assistance (ERA) focuses on one of the most time‑consuming parts of modern science: writing and iterating computational experiments.

ERA uses Gemini models to generate and optimize scientific code for simulations and experiments. The system was described in a publication in Nature and helped build a prototype platform called Computational Discovery, which is now available through a trusted tester program in Google Labs.

By automating parts of scientific programming and experimentation, ERA allows researchers to iterate through potential experiments more quickly than manual workflows typically allow.

NotebookLM‑Based Research Tools

Gemini for Science also incorporates NotebookLM‑based workflows, enabling researchers to synthesize large volumes of scholarly literature and extract structured insights from scientific papers.

These tools help scientists analyze thousands of papers, identify emerging research themes, and generate research artifacts grounded directly in source material.

How Google Labs and Antigravity Fit In

Google positioned Gemini for Science as part of its experimental AI ecosystem.

Three primary prototype tools are available through Google Labs, where researchers can explore and test the early capabilities of the system.

At the same time, the platform connects to Google Antigravity, Google’s agent‑first development platform designed to move AI from simple prompts toward systems that can take action autonomously.

Through Antigravity’s Science Skills, AI agents can connect to scientific datasets and tools, allowing them to retrieve information, run analyses, and generate research outputs as part of a larger automated workflow.

Integration With Life‑Science Databases

One of the more ambitious aspects of the launch is its planned data connectivity.

Google says the Science Skills layer can link Antigravity‑powered agents to more than 30 major life‑science databases and tools, enabling direct access to scientific datasets and research resources.

The company has not publicly specified which databases are included or how the integrations operate, but the concept is to allow AI systems to interact directly with scientific knowledge sources rather than relying solely on static training data.

ERA’s Early Real‑World Impact

Among the tools in the suite, ERA already has documented research results.

According to Google Research, the system:

• Was described in a Nature publication describing its capabilities in scientific coding
• Helped build the Computational Discovery prototype
• Is currently being tested through a Google Labs trusted tester program

These early deployments suggest that AI‑assisted experimentation and code generation could become a practical part of research workflows.

Part of Google’s Shift to Agentic AI

Gemini for Science also reflects a broader shift across Google’s AI strategy.

At I/O 2026, the company emphasized moving from AI systems that simply respond to prompts toward agentic AI systems that can perform tasks and interact with external tools.

Gemini for Science embodies this idea by packaging multiple specialized agents—hypothesis generation, algorithm search, and experimental coding—into a coordinated research environment.

Instead of acting as a single chatbot, the platform represents a network of AI tools designed to collaborate with scientists and accelerate the pace of discovery.

If the approach succeeds, it could significantly change how computational research is conducted, especially in fields such as biology, medicine, and complex systems science.