Claude Science: AI Workbench for Scientists Launches in Beta
Claude Science, launched June 30, 2026, replaces the fragmented researcher workflow of switching between PubMed, Jupyter, R, and dozens of databases with a single environment. Every output it produces is auditable and reproducible. Available in beta to Pro, Max, Team, and Enterprise users on macOS and Linux.
Claude Science gives you one place to do all of your research.
Scientific work is fragmented: PubMed for literature, Jupyter for code, R for stats, a cluster terminal for big jobs, and a different web interface for each of the dozens of databases you query. Claude Science brings all of that into one app. You analyze literature, run experiments, generate figures, and write manuscripts without switching tools.
It launched June 30, 2026 in beta, available on macOS and Linux for Claude Pro, Max, Team, and Enterprise users. Team and Enterprise admins must enable it first.
Every output carries a full audit trail.
When Claude Science generates a figure, it attaches four things: the exact code that produced it, the environment (package versions and dependencies), a plain-language description of how it was made, and the full message history. A collaborator can open any result six months later and reproduce it exactly.
You can revise figures in plain language — "change the y-axis to log scale" or "remove the gridlines" — and Claude Science edits its own code to match. A reviewer agent runs throughout every session, checking citations, flagging numbers that can't be traced to source data, and correcting figures that don't match their underlying code before you see the output.
60+ domain-ready tools, one conversation.
A generalist coordinating agent backs every Claude Science session with over 60 curated skills and connectors, pre-configured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. When you ask a question in plain language, specialist agents query and synthesize across sources like UniProt, PDB, Ensembl, ClinVar, and ChEMBL — you describe what you want; the agents handle the database querying.
Claude Science also connects to NVIDIA's BioNeMo Agent Toolkit, giving you native access to Evo 2, Boltz-2, and OpenFold3 in the same conversation. Scientists can save any pipeline as a reusable skill or add a connector to their lab's own tools, with future sessions inheriting them automatically.
Compute that scales to your lab's infrastructure.
Claude Science runs where your data already lives — your laptop, a Linux box, or an HPC login node over SSH. Large or sensitive datasets never leave your systems. For on-demand scaling, connect a Modal account and the agent will scale an analysis from a single GPU to hundreds as needed.
When an analysis needs cluster resources, Claude Science drafts a plan, asks for your approval before touching any new resource, and submits the job itself. Because agents work inside a running session that holds context in memory, even massive datasets only need to be loaded once. You can fork the session at any point to compare two approaches without losing the original thread.
What scientists have done in beta: Manifold Bio used it for end-to-end target nomination — assessing surface expression, trafficking, and safety across tissues with proprietary context built in. Jérôme Lecoq at the Allen Institute used it to build a 20-skill computational review pipeline; what used to take his team up to two years now produces 10+ reviews, many over 100 pages, with citations checked by reviewer agents. Stephen Francis at UCSF used it for glioma germline workups, completing comprehensive multi-approach analyses in roughly one-tenth the previous time.
What you can do now: access Claude Science at claude.com/science on macOS or Linux with a Pro, Max, Team, or Enterprise plan. Every artifact it generates includes the code, environment, and message history needed to reproduce it. Describe what you want in plain language — pull data from UniProt or PDB, run a genomics pipeline, refine a figure. The reviewer agent checks citations and calculations automatically. For intensive compute, connect your HPC cluster or a Modal account and the agent scales the job for you.
Anthropic is also supporting up to 50 AI for Science projects with up to $30,000 in credits each, with applications open through July 15, 2026.