Anthropic launched Claude Science in beta on June 30, 2026, packaging its existing Claude models into a scientific workbench with over 60 curated skills and connectors across genomics, proteomics, structural biology, and cheminformatics. The app integrates literature analysis, code execution, figure generation, and compute orchestration, with reviewer agents that check citations and calculations - a bet on workflow reproducibility rather than a new biology model.
Claude Science uses the same Claude models rather than a specialized biology model. Its differentiation comes from the surrounding system: a coordinating agent, specialist agents, access to scientific databases such as UniProt, PDB, and ChEMBL, and native support for artifacts like protein structures and genome browser tracks. The app runs on macOS, Linux, remote machines, SSH, and HPC login nodes, placing it where researchers already work.
What Claude Science includes
The beta gives Pro, Max, Team, and Enterprise users one environment for multi-step research workflows. Anthropic says the app can connect to resources including Ensembl, Reactome, ClinVar, GEO, and NVIDIA BioNeMo tooling, alongside lab-specific pipelines. A reviewer agent checks citations and calculations, and the system preserves an auditable history for each result, aiming to make outputs reproducible enough for team review.
The product fits into the broader category of AI for Science & Research, but its specific promise is workflow control. It is closer to a managed research environment than a chatbot wrapper, with local or lab-controlled compute, reusable skills, and evidence checking built into the product surface.
For computational biology and data science teams
The core evaluation question is whether Claude Science can deliver reproducible analysis. Teams should test whether outputs can be rebuilt from stored code and environment, whether source citations survive adversarial review, and whether sensitive datasets remain inside approved compute boundaries. The claims worth testing are provenance, citation fidelity, and controlled data handling - not autonomous drug discovery.
Anthropic also plans to pursue its own drug programs for neglected diseases. A demonstration cited in the launch analyzed 100 rare genetic diseases in under an hour and flagged 32 candidates for computational screening. Those are screening claims, not clinical evidence. Wet-lab validation, toxicity work, and clinical trials remain the bottleneck.
Why this matters for science and research professionals
Claude Science matters not because it introduces a new biology model, but because it packages existing AI into a reproducible research workbench with scientific connectors, compute access, and reviewer agents. For drug-discovery and bioinformatics teams, the immediate value lies in faster literature triage and early candidate screening. The credibility test will be independent validation from labs using the tool on published, reproducible workflows, and documented governance around audit trails and data retention for regulated or patient-linked work.
Your membership also unlocks:
