Explore
Search literature and databases, inspect files, and form source-linked candidate questions.
Independent guide · checked August 2, 2026
Quick answer
Claude Science joins analysis to its audit trail.
Use it when a scientist needs literature, code, scientific artifacts, and compute in one beta workbench. Preserve source, environment, data, reviewer, and external-validation evidence rather than treating agent self-checks as proof.
Verified product record
Product
Claude Science is Anthropic's beta AI workbench for scientists.
Published plans
Pro, Max, Team, and Enterprise; Team and Enterprise users need an administrator to enable it.
Systems
The app is available on macOS and Linux and can work locally, over SSH, or from an HPC login node.
Scientific setup
A coordinating agent has more than 60 curated skills and connectors across named scientific domains.
Artifacts
Figures and manuscripts include generating code, environment, plain-language description, and message history.
Compute
Can plan jobs for a lab's own infrastructure or supported on-demand compute and asks before reaching new resources.
Flowith relationship
Independent editorial guide; no embedded Flowith integration is claimed.
Evidence loop
Explore, execute, audit, validate
Execute
Run versioned code and environments locally, over SSH, on HPC, or through approved compute.
Audit
Keep artifacts, code, environment, inputs, history, citations, calculations, and reviewer findings together.
Validate
Use independent statistics, domain review, controlled experiments, safety checks, and external replication.
Continue by decision intent
Official source
Anthropic: Claude Science, an AI workbench for scientists — beta access, plans, systems, skills, artifacts, compute, reviewer agents, workflows, and current data boundary.
Claude Science questions, answered
Claude Science is Anthropic's beta AI workbench for scientific research. It combines literature and multi-step analysis, curated scientific skills and connectors, rich artifacts, code and environment history, compute access, and reviewer agents.
Anthropic's launch lists Pro, Max, Team, and Enterprise plans. Team and Enterprise users need an administrator to enable it. Verify live plan, region, seat, price, limits, terms, and beta state.
Anthropic lists macOS and Linux. The app can run in a scientist's local environment or connect over SSH or through an HPC login node, subject to local infrastructure and policy.
Anthropic says large or sensitive datasets can remain on the lab's own infrastructure and only context needed for each analysis step is sent to Claude. That still requires a documented data-flow, minimization, access, retention, contract, and policy review.
No. A reviewer agent can flag citations, calculations, traceability, or figure-code mismatches, but it is part of the same product workflow. Independent domain, statistical, code, data, safety, and experimental validation remain necessary.
This is an independent evaluation page. Anthropic's product availability does not establish a Flowith integration, so verify each product separately.