Anthropic launches Claude Science, an AI workbench for labs

Anthropic's pitch to lab scientists is that most research days get eaten by shuffling between databases, file formats, and one-off viewers, and that a single AI environment can absorb that shuffle. Claude Science, now in beta, is the company's attempt to package literature analysis, multi-step research, code execution and figure generation into one workbench with what Anthropic calls an auditable history of how each output was made.

The product ships with over 60 curated skills across genomics, single-cell analysis, proteomics, structural biology and cheminformatics. It connects to reference resources including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL and GEO, and integrates with NVIDIA's BioNeMo Agent Toolkit and models such as Evo 2, Boltz-2 and OpenFold3. On screen, Claude Science renders 3D protein structures, genome browser tracks and chemical structures alongside the code that produced them.

Anthropic's evidence comes from named labs. Jérôme Lecoq at the Allen Institute is quoted saying a review workflow that could take his team as many as two years is now materially faster, and Stephen Francis at the UCSF Brain Tumor Center says the app accelerated germline workups to roughly one-tenth the time they previously took. Take those numbers as reported, not settled: they are single-lab testimonials from launch-partner researchers, not independent benchmarks, and the announcement does not describe how the built-in reviewer agent behaves when it is wrong about a citation.

Beta access is available on macOS and Linux for Pro, Max, Team and Enterprise subscribers, with a discounted Team plan for academic and nonprofit labs. Anthropic is also funding up to 50 AI for Science projects with up to $30,000 in Claude credits and up to $2,000 in Modal compute credits each, with applications open through July 15, 2026 and a project window running September 1 to December 1, 2026. The bet worth watching is not that Claude gets better at biology, but that the workflow glue itself becomes the product: if researchers who currently splice together Jupyter, R and half a dozen portals live inside this instead, the integrations become the moat.