Anthropic launches Claude Science: an AI workbench for scientists [2026]

Anthropic today launched Claude Science, an AI workbench designed specifically for scientists and researchers. The announcement came with the release of Claude Sonnet 5, solidifying the day as one of the company’s biggest in 2026.

Available in public beta for subscribers of Pro, Max, Team, and Enterprise plans, Claude Science promises to unify the fragmented ecosystem of scientific research tools into a single AI-controlled environment.

The Problem Claude Science Solves

Scientific researchers deal with dozens of data sources — each with its own schema and query language — file formats requiring specific pipelines and viewers, and an arsenal of tools including PubMed, Jupyter, R, and cluster terminals. Moving between these systems consumes time and introduces error.

Claude Science consolidates these tools into a single environment where scientists can conduct all stages of their work: from literature analysis to executing multi-step research, through the production of artifacts, figures, and publication-ready manuscripts.

How It Works

The user interacts with a generalist coordinating agent that has access to over 60 pre-configured skills and connectors for areas such as:

Genomics
Single-cell
Proteomics
Structural biology
Computational chemistry

These agents can create specialized sub-agents and interact with agents created by users themselves. A dedicated reviewer agent checks citations and calculations, automatically flagging and correcting errors.

Rich and Fully Reproducible Scientific Artifacts

Claude Science generates figures and manuscripts along with the code that produced them. It natively renders complex scientific artifacts, including:

3D protein structures
Genome browser tracks
Chemical structures

Each generated figure comes with the exact code, environment description, a natural language explanation of how it was created, and the full conversation history. The user can request changes in natural language — “remove grid lines” or “change the axis to log scale” — and the agent edits its own code.

Intelligent Compute Management

Large-scale analyses — such as protein folding or genomic pipelines on massive datasets — require cluster job setup, completion verification, and results collection. Claude Science automates this process:

Elaborates a plan
Requests authorization before accessing new resources
Allows reviewing or revoking any decision
Submits the job to the computing resources the lab already uses (HPC via SSH or Modal for on-demand computing)
Scales from a single GPU to hundreds as needed

Since agents work within an active session that maintains context in memory, massive datasets only need to be loaded once. And everything runs on the lab’s own infrastructure — laptop, Linux server, or HPC login node — without sensitive data needing to leave the systems where it already resides.

Connectivity with Scientific Ecosystems

Claude Science uses the skills from the NVIDIA BioNeMo Agent Toolkit to natively connect to models and libraries such as Evo 2, Boltz-2, and OpenFold3. Researchers can also connect their own models, datasets, and pipelines, saving any workflow as a reusable skill.

Real Use Cases (Closed Beta)

During the months of closed beta, Claude Science was tested by research institutions with significant results:

Manifold Bio — a startup developing tissue-targeted drugs — used Claude Science to name targets for its latest experiments. For each tissue and target, the system evaluated surface expression, trafficking, and safety, ranking candidates against the criteria developed by the company. The differentiator, according to Manifold, was the ability to do this end-to-end.

Jérôme Lecoq, a neuroscientist at the Allen Institute, used Claude Science to build a “computational review template” with about 20 custom skills. The system reads thousands of articles, extracts central theses and quantitative findings, and produces reviews over 100 pages long. What previously took up to two years to write is now done in a fraction of the time — Lecoq has already produced about 10 reviews.

Stephen Francis, associate professor and epidemiologist at the UCSF Brain Tumor Center, used Claude Science for studies on the molecular epidemiology of glioma. His lab was able to perform comprehensive germline analyses in approximately one-tenth of the previous time, with independently validated results.

Availability and Pricing

Plans: Pro, Max, Team, and Enterprise (Teams and Enterprises require admin enablement)
Platforms: macOS and Linux (local or remote via SSH/HPC)
Academic Discount: Team plans with special pricing for active labs in academic institutions and non-profit research organizations
Support Program: Up to 50 “AI for Science” projects with up to US$ 30,000 in credits (Modal offers up to US$ 2,000 in additional compute for selected projects)
Applications open until July 15, 2026, with notification by July 31. Projects run from September 1 to December 1, 2026.