Anthropic Launches Claude Science, an AI Workbench for Scientific Research

Anthropic dropped Claude Science today, an AI workbench built for scientists. The beta launch covers Pro, Max, Team, and Enterprise users, bundling the tools researchers already lean on, producing artifacts you can trace back to their source, and handing flexible access to compute power.

A Single Home for Messy Workflows

Science has a fragmentation problem. Databases like PubMed, UniProt, PDB, and Ensembl each speak their own query language. Claude Science sweeps them into one space where researchers can read literature, run multi-step analyses, and polish figures and manuscripts until they're submission-ready.

You talk to a generalist coordinating agent that taps into more than 60 curated skills and connectors, pre-tuned for genomics, single-cell work, proteomics, structural biology, and cheminformatics. That agent can spin up specialist agents you've built yourself. A dedicated reviewer checks citations and calculations, flagging mistakes and fixing them on the spot.

Reproducibility, Built In

The platform renders rich scientific outputs natively, 3D protein structures, genome browser tracks, chemical structures. When it generates a figure, it packages the exact code and environment that made it, a plain-language explanation of how it got there, and the full conversation history. Months later, you can come back and validate or reproduce the result. Want to drop gridlines or switch an axis to log scale? Say it in plain English, and the agent edits its own code.

Compute Without the Hassle

Large analyses, protein folding, genomics pipelines over massive datasets, often turn researchers into job schedulers. Claude Science takes over: it drafts a plan, asks before grabbing new resources, and lets you review or veto any decision before it submits jobs to your existing infrastructure, whether that's an HPC cluster over SSH or a Modal account for on-demand compute. It scales from a single GPU to hundreds as needed.

Because agents work inside a live session that keeps context in memory, even giant datasets load once. The platform runs on your lab's own hardware, so sensitive data stays put. As the pipeline churns, the reviewer agent checks outputs, flagging bad citations, numbers without provenance, and figures that don't match their code.

Ready for the Lab on Day One

Claude Science taps into NVIDIA's BioNeMo Agent Toolkit for native access to life sciences models and libraries, Evo 2, Boltz-2, OpenFold3. You can save custom pipelines as reusable skills or plug in your lab's favorite tools through connectors; future sessions pick them up automatically.

Early Wins from Beta Users

Anthropic shared a few early stories. Manifold Bio, a company designing tissue-targeting medicines, used Claude Science to nominate targets for its latest experiments, assessing surface expression, trafficking, and safety across multiple criteria. The firm said the platform handled it end-to-end, gathering the right data and applying judgment with built-in context from past programs.

Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent computational review template with about 20 custom skills. Sub-agents read through thousands of papers, pull central claims and key quantitative findings, and store them in an evidence state database. A key component is the use of actor-critic pairs: one agent creates content while a separate reviewer agent evaluates it for accuracy and citation fidelity. Lecoq reported that before Claude Science, writing such a review could take up to two years; he now has about ten reviews, many exceeding 100 pages, with citations checked by reviewer agents.

Stephen Francis, an associate professor and epidemiologist at the UCSF Brain Tumor Center, used Claude Science to support studies on the molecular epidemiology of glioma. He said the platform accelerated comprehensive germline workups across multiple approaches to roughly one-tenth the previous time, and his group independently validated the results, confirming rapid and robust analyses.

Pricing and Support

Claude Science is available in beta on macOS and Linux for Pro, Max, Team, and Enterprise plans. Team and Enterprise users need their admin to enable the app. Anthropic now offers a Team plan with discounted seats for active scientific labs at academic institutions and nonprofit research organizations.

Anthropic is also supporting up to 50 Claude Science AI for Science projects, with up to $30,000 in credits. Modal will provide up to $2,000 in compute for select projects. Applications are open through July 15, 2026, with award notifications by July 31. Projects will run from September 1 to December 1, 2026. Researchers can join the AI for Science Discourse community for updates and feedback.