Claude Science is Anthropic's attempt to turn AI research support into a workbench, not another chat surface. The beta brings scientific tools, curated skills, connectors, compute access, artifacts, and review into one environment.
The release targets a familiar research drag: databases with different schemas, file formats that need custom viewers, notebooks, terminals, figures, manuscripts, and compute jobs spread across too many places.
The workbench absorbs research tool switching
Anthropic says Claude Science integrates tools and packages researchers use, including preconfigured support for genomics, single-cell, proteomics, structural biology, cheminformatics, and more. Users work with a coordinating agent that can call specialist agents and use over 60 curated skills and connectors.
The product also runs where researchers already work: local machines, Linux systems, remote machines over SSH, or HPC login nodes. That matters for labs that cannot casually move large or sensitive datasets out of their current infrastructure.
Artifacts make the system reviewable
The strongest part of the release is artifact discipline. Anthropic says Claude Science produces figures, manuscripts, code, environment details, plain-language generation history, and message history so outputs can be inspected later.
It also includes a reviewer agent that checks citations, calculations, untraceable numbers, and figures that do not match the code. That does not remove scientific review. It gives researchers a clearer trail for review, reproduction, and correction.
Related services: Knowledge, Quality, Claude workspace expertise