Anthropic launches Claude Science AI workbench for researchers
The new application integrates databases, computing resources, and specialized tools into a single environment with built-in reproducibility tracking.
Anthropic has released Claude Science, an AI-powered application designed to consolidate the fragmented toolkit scientists navigate daily into a unified research environment. The company announced the beta launch for Claude Pro, Max, Team, and Enterprise subscribers.
The workbench addresses a persistent friction point in scientific research: the need to switch between dozens of specialized databases, file formats, and tools—from PubMed to Jupyter notebooks to high-performance computing clusters—each with distinct interfaces and workflows.
How the system works
Claude Science operates through a coordinating agent with access to more than 60 pre-configured skills spanning genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. The system can generate specialist sub-agents for specific tasks and includes a reviewer agent that validates citations and calculations in real time.
The application runs on existing infrastructure—whether a researcher's laptop, Linux workstation, or HPC login node—meaning sensitive datasets never leave local systems. Only the contextual information needed for each analysis step is transmitted to Claude's servers.
A distinguishing feature is the platform's approach to reproducibility. Every figure, manuscript section, or analysis output includes the exact code that produced it, the computational environment specifications, a plain-language explanation of the methodology, and the complete message history. Researchers can request modifications in natural language, and the system edits its own code accordingly.
Compute management and domain integration
For computationally intensive tasks like protein folding or large-scale genomics pipelines, Claude Science handles job submission to computing resources automatically. It drafts execution plans, requests permission before accessing new resources, and scales from single GPUs to hundreds as needed through integrations with HPC clusters or cloud platforms like Modal.
The system connects to specialized life sciences resources through NVIDIA's BioNeMo Agent Toolkit, providing native access to models including Evo 2, Boltz-2, and OpenFold3. It also queries domain-specific databases—UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and others—synthesizing information across sources that typically require separate navigation.
Early adoption results
Manifold Bio used Claude Science to identify targets for tissue-specific drug candidates, with the system assessing surface expression, trafficking, and safety profiles while incorporating the company's proprietary historical data. The company highlighted the platform's ability to complete end-to-end analysis rather than serving as a simple coding assistant.
Neuroscientist Jérôme Lecoq at the Allen Institute built a multi-agent system using approximately 20 custom skills to automate literature review writing. His workflow processes thousands of papers, extracts quantitative findings, constructs narrative structures, and generates cross-study figures—reducing what previously took two years to a fraction of that time.
Stephen Francis, an epidemiologist at the UCSF Brain Tumor Center, reported that Claude Science accelerated his lab's germline variant analysis for glioma research to one-tenth the previous timeline while maintaining independently validated accuracy.
Why it matters
Scientific productivity often bottlenecks not at the hypothesis stage but in the mechanical work of data wrangling, tool switching, and documentation. By consolidating these steps and embedding reproducibility from the start, Claude Science could lower barriers to complex analyses—particularly for smaller labs without dedicated bioinformatics staff. The built-in audit trail also addresses a growing concern in computational research: the difficulty of reproducing published results when code, environments, and intermediate steps aren't systematically preserved.
Anthropic is offering up to 50 research projects $30,000 in credits each, with applications open through July 15, 2026. Modal will provide an additional $2,000 in compute resources for selected projects.
Details were first reported by Anthropic in its product announcement.
This is an original analysis by the Omega editorial team. Source reporting: The Verge.
Want systems like this working for your business?
Book a Call