Saturday, Oct 17 | --:--
Back to home

Claude Science: Vertical Workbench Bet—Workflow and Auditability, Not a New Base Model

Anthropic introduced Claude Science beta—an AI research workbench with 60+ skills, auditable artifacts, and HPC/Modal compute. Wins scientists on workflow, not another frontier GA.

Times of AI Desk 5 min read San Francisco, CA View as Markdown
Cover illustration for Claude Science: Vertical Workbench Bet—Workflow and Auditability, Not a New Base Model

Claude Science brings research tools, compute access and traceable outputs into one workbench. The launch describes a scientific workflow product rather than a new base model.

Anthropic introduced Claude Science, an AI workbench for scientists, in beta for Claude Pro, Max, Team, and Enterprise. The app integrates common research tools and packages, produces auditable figures and manuscripts, runs on local macOS/Linux or remote SSH/HPC, and manages lab compute (HPC over SSH or Modal on demand). A coordinating agent uses 60+ curated skills (genomics, single-cell, proteomics, structural biology, cheminformatics, etc.), can spawn specialists, and includes a reviewer agent for citations and calculations.

What’s in the workbench

From Anthropic’s primary post:

Capability Detail
Environment Single research app vs fragmented PubMed / Jupyter / R / cluster terminal
Artifacts Figures + manuscripts with code, environment, plain-language history
Rendering Native 3D proteins, genome tracks, chemical structures
Compute Plan → approve → submit jobs on lab HPC or Modal; scale GPU count
Data locality Runs on lab machines; datasets can stay on lab systems; required analysis context is sent to Claude
Skills 60+ preconfigured; NVIDIA BioNeMo toolkit (Evo 2, Boltz-2, OpenFold3, etc.)
Plans Beta: Pro, Max, Team, Enterprise (Team/Enterprise need admin enable)
Programs Up to 50 AI-for-Science projects with up to $30K credits; Modal up to $2K compute; apps through Jul 15, 2026

Anthropic also pointed to discounted Team seats for academic/nonprofit research labs. Early user anecdotes in the post include Manifold Bio, Allen Institute, and UCSF Brain Tumor Center workflows.

Limits

  • Beta only; Free-tier users are out.
  • Team/Enterprise need admin enable—not automatic seat entitlement.
  • AI-for-Science credit program capped (50 projects / $30K; Modal $2K) with July 15 application cutoff.

Correction (October 14, 2026): Clarified that analysis context is sent to Claude even when datasets remain on lab infrastructure; removed unrelated launch-day context.

Sources

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading