Anthropic launched Claude Science, a customizable AI workbench integrating 60+ curated scientific skills, automated compute management, and reproducible artifact generation — a significant step toward AI-native scientific research.
Replaces PubMed, Jupyter, R, and cluster terminals with a single unified research interface.
Every output includes code, environment, and message history for full traceability.
Automates job submission to local, HPC, or cloud GPUs — scales from one to hundreds.
Actor-critic pairs: specialist agents create, reviewer agent validates accuracy and citations.
Claude Science consolidates the tools scientists use daily — PubMed, Jupyter, R, cluster terminals, and dozens of specialized databases — into a single interface. Researchers interact with a generalist coordinating agent with access to over 60 curated skills and connectors for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.
When Claude Science generates a figure, it includes the exact code and environment that produced it, a plain-language description of how it was created, and the full message history. Users can request edits in plain language — "remove gridlines" or "change the axis to log scale" — and the agent edits its own code.
Natively renders 3D protein structures, genome browser tracks, chemical structures, and other scientific visualizations — with full audit trails for validation and reproducibility.
Researchers no longer context-switch to infrastructure — Claude Science handles job submission, monitoring, and result retrieval end-to-end.
A coordinating generalist agent delegates work to specialist sub-agents. The key architectural innovation is the actor-critic pair: one agent creates content while a separate reviewer agent evaluates it for accuracy and citation fidelity.
UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, GEO — pre-configured with native connections.
Evo 2 (genomics), Boltz-2 (protein structure), OpenFold3 — accessed via BioNeMo Agent Toolkit.
Save any pipeline as a reusable skill; future sessions inherit custom tools automatically.
Over 60 curated skills and connectors for genomics, single-cell, proteomics, structural biology, and cheminformatics.
Target nomination for tissue-targeting medicines — end-to-end assessment of surface expression, trafficking, and safety in a fraction of previous time.
Jerome Lecoq built a 20-skill multi-agent review template. Cut long-form review writing from 2 years to months — 10+ reviews exceeding 100 pages each.
Stephen Francis accelerated glioma molecular epidemiology analysis by ~10x, with independently validated results.
Unlike general coding assistants, Claude Science handles the entire workflow end-to-end with built-in domain context.
Test against your workflow — Identify one ongoing analysis (scRNA-seq, protein prediction, literature review) and benchmark against current methods.
Apply for funding — Anthropic's AI for Science program (through July 15, 2026) offers up to $30,000 in credits plus $2,000 in Modal compute.
Test skill creation — The highest-leverage feature: can it integrate with your existing validated pipelines without workflow redesign?
Join the community — ai4science.discourse.group surfaces real-world issues, benchmarks, and patterns faster than docs.
Tier guide: Tier 1 = Primary institutional source (official announcement) · Tier 2 = Company-owned media or trade publication · Tier 3 = Breaking news / single-source — verify independently
Sources: Anthropic official announcement. Article content extracted as markdown. Generated July 3, 2026 by Gambit (Hermes Agent).