Claude Science AI Workbench Release

Anthropic has introduced Claude Science, an AI-powered workbench designed to consolidate the fragmented tools and data pipelines used in scientific research into a single environment. The platform aims to accelerate discovery and healthcare interventions by automating tedious data synthesis, managing complex compute resources, and ensuring that all scientific outputs are fully reproducible.

Integrated Research Environment and Reproducibility

Claude Science provides a unified workspace where researchers can analyze literature, execute multi-step research, and refine manuscripts and figures. To ensure scientific integrity, every output includes an auditable history, including the exact code and environment used to produce the result, a plain-language description of the process, and the full message history.

Key capabilities for scientific visualization include:

Native Rendering: The platform natively displays 3D protein structures, chemical structures, and genome browser tracks.
Iterative Refinement: Users can edit figures using plain language (e.g., requesting a log scale for an axis), and the agent updates the underlying code automatically.
Flexible Deployment: The workbench can be accessed locally on macOS and Linux, or on remote machines via SSH or HPC login nodes.

Scalable Compute and Data Management

Claude Science automates the submission and management of heavy computational jobs, such as protein folding or genomics pipelines. The system drafts a plan and requests user approval before accessing new resources, allowing researchers to scale from a single GPU to hundreds as needed.

Compute Infrastructure

Hybrid Execution: It integrates with existing lab infrastructure (laptops, Linux boxes, HPC clusters via SSH) and supports compute-on-demand via Modal.
Data Privacy: Large or sensitive datasets remain on the lab's own systems; only the necessary context for each analysis step is sent to Claude.
Session Persistence: Agents operate within a running session that holds context in memory, meaning massive datasets only need to be loaded once.

Domain-Specific Intelligence and Tooling

Claude Science is pre-configured for genomics, single-cell, proteomics, and cheminformatics, utilizing a coordinating agent with access to over 60 curated skills and connectors.

Integration with Scientific Ecosystems

Database Synthesis: Specialist agents can query and synthesize data across diverse sources, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO.
BioNeMo Integration: The platform uses NVIDIA’s BioNeMo Agent Toolkit to connect to life sciences models and libraries, including OpenFold3, Boltz-2, and Evo 2.
Customizability: Researchers can save their own pipelines as reusable skills or create connectors for proprietary lab tools, which are then inherited by future sessions.

Real-World Research Applications

Beta users have applied Claude Science to various high-impact scientific tasks, including CRISPR screen design and single-cell RNA sequencing analysis.

Drug Target Nomination: Manifold Bio used the platform to nominate targets for experiments by assessing surface expression, trafficking, and safety against proprietary internal data.
Automated Literature Review: A neuroscientist at the Allen Institute developed a multi-agent "computational review template" using 20 custom skills to synthesize thousands of papers into long-form reviews. This process utilized actor-critic pairs, where one agent creates content and a reviewer agent validates accuracy and citation fidelity.
Epidemiological Analysis: At the UCSF Brain Tumor Center, the platform was used to support studies on the molecular epidemiology of glioma, reducing the time required for comprehensive germline workups to roughly one-tenth of previous timelines.

Availability and Support

Claude Science is currently in beta for Claude Pro, Max, Team, and Enterprise users on macOS and Linux. Anthropic is offering discounted Team plan seats for active scientific labs at academic institutions and nonprofit research organizations.

Additionally, Anthropic is supporting up to 50 "AI for Science" projects with up to $30,000 in credits, with select projects receiving up to $2,000 in compute from Modal. Applications for these projects are open through July 15, 2026.