Claude Science
Also known as: Claude for Science, Anthropic Science workbench
Researchers typically bounce between PubMed for literature, Jupyter notebooks for code, R for statistics, and a cluster terminal for compute. Claude Science is Anthropic's attempt to collapse that stack: a desktop application for macOS and Linux that connects Claude to more than 60 scientific databases and toolkits spanning genomics, proteomics, structural biology, and cheminformatics, all within a single persistent project workspace.
The architecture is explicitly multi-agent. A primary coordinating agent oversees the overall research task, either spawning specialist sub-agents to divide the workload or routing to custom agents a researcher has built for their own use cases. A separate reviewer agent independently audits citations and calculations before output surfaces, catching and correcting errors as it goes. Every result is traceable back to the code that produced it.
Claude Science is not a new or different model: it runs standard Claude models available on Pro, Max, Team, and Enterprise plans. The product is better understood as a domain-specific harness that applies agentic workflow patterns to scientific research. For builders, it's a concrete example of how 'AI workbench' products are being constructed: not new intelligence, but new scaffolding that structures what the model does and how it verifies its own outputs.