Claude Science: An AI Workbench for Researchers

Claude Science, which people also search for as Claude for Science, is an app Anthropic released in beta on June 30, 2026. It gathers a researcher's literature search, analysis, figures, and compute in one workbench, and it runs on macOS and Linux for Claude Pro, Max, Team, and Enterprise plans.

What is Claude Science and what does it replace?

The data and tools one project needs sit in separate places, and the app pulls them into one environment instead of adding another model. A single project can draw on UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO, and each of those databases has its own schema and query language. Claude Science is set up for genomics, single-cell, proteomics, and cheminformatics, with more than 60 scientific databases behind it.

Researchers keep working where they already work. Like a Jupyter Notebook, the app runs locally on macOS or Linux, or on a remote machine reached over SSH or an HPC login node. For a lab that today switches between a literature search, a notebook, and a terminal on a cluster, that means one place for all three.

How Claude Science keeps every figure reproducible

A co-author or reviewer can ask about a figure months later, and the record that made it is still attached. When the app generates a figure, it includes the exact code and environment that produced it, a plain-language description of how it was created, and the full message history. The product page says the same about tables and notebooks: each can still be reproduced months later, and edited or defended when needed.

The app also renders 3D protein structures and genome browser tracks natively, along with chemical structures. A structural biologist can check a model in the same window that produced it, with no separate viewer.

Claude Science agents: coordinator, specialists, and reviewer

Work is split across three layers of agents, so one request does not depend on a single model call. Researchers talk to a generalist coordinating agent that has over 60 curated skills and connectors set up for genomics, single-cell, proteomics, structural biology, cheminformatics, and more. The coordinator can spin up other agents and engage specialist agents that users create.

The three layers of agents

The reviewer runs in the background while the other agents work. It flags incorrect citations, numbers that cannot be traced back to their source, and figures that do not match the code underneath them. For a lab writing a long manuscript with many references, that covers the three errors a quick read of a draft is most likely to miss.

NVIDIA BioNeMo and compute from one GPU to hundreds

Models such as Evo 2, Boltz-2, and OpenFold3 are reached through skills, so a lab does not wire each one up by hand. Claude Science uses the skills in NVIDIA's BioNeMo Agent Toolkit to connect natively to the life sciences models and libraries in BioNeMo. A protein-folding run with Boltz-2 then starts from the same conversation as the rest of the project.

Heavy jobs go to compute the lab already uses. The app drafts a plan, asks before reaching new resources, and lets the user review or revoke any decision before it writes and submits the job.

Where compute comes from

The analysis can scale from a single GPU to hundreds as needed. Because the app runs on hardware the lab already controls, large or sensitive datasets do not have to leave those systems, and only the context needed for each step is sent to Claude.

Three labs report faster reviews, workups, and target picks

Each report comes from a named lab and describes one task, so none of them is a controlled study. Anthropic's announcement names Manifold Bio, the Allen Institute, and the UCSF Brain Tumor Center.

What each lab reports

At the Allen Institute, Jérôme Lecoq built a multi-agent computational review template from about 20 custom skills. Before Claude Science, a review could take his team as many as two years, and he now has about 10 reviews, with citations checked by reviewer agents. At the UCSF Brain Tumor Center, Stephen Francis reports full germline workups across multiple approaches in roughly one-tenth the time they took before.

Who can use the Claude Science beta and where to get it

Access depends on the plan and the operating system, and on Team and Enterprise it also depends on an admin. Anthropic released the beta for Claude Pro, Max, Team, and Enterprise users and says it will keep refining the app as feedback comes in. The announcement names macOS and Linux only.

Access at a glance

A member of a Team or Enterprise plan asks the admin to switch Claude Science on. The announcement publishes no separate Claude Science price on top of the plans, and its only dollar figures belong to the AI for Science program.

For how the Claude plans themselves compare, see What Is Claude AI? Free Plan, Models, vs ChatGPT.

Academic discounts and the AI for Science program

Labs at universities and nonprofit research organizations have two routes to lower costs, a discounted Team plan and a credit program. Anthropic offers a Team plan with discounted seats for active scientific labs at academic institutions and nonprofit research organizations. The product page adds that scientists at these institutions around the world can get Claude at no cost to start.

AI for Science program

Award notifications were due by July 31, so the funded round is already underway and new applicants cannot join it. A lab that missed the window can still use the discounted Team plan.