Claude Science
Anthropic's AI workbench for scientists — the research analogue of Claude Code, aimed at end-to-end scientific workflows for pharma and biotech.
1. Core Product / Service
Claude Science is anthropic|Anthropic's science-focused product, launched June 30, 2026 [1]. It is positioned to scientific research what Claude Code does for software engineering: a workbench that organizes research projects, connects to data sources, and delegates sub-tasks to specialized assistants [2].
Local research (2026-09) surfaced the product architecture:
- Connectors — integrations to scientific data sources; "featured connectors" ship with public/free APIs, while others require the user to bind their own account.
- Skills — packaged capabilities (e.g. ESM2 for protein language modeling); some require user-provisioned compute.
- Kernels — compute/execution environments attached to the workbench.
- Reviewer system — a built-in critique/review mechanism layered over research outputs.
2. Target Users & Pain Points
- Pharma/biotech researchers — the primary launch audience per STAT [3].
- Academic labs wanting structured, reproducible research workflows.
- Pain solved: general coding assistants lack domain context and end-to-end workflow plumbing; Claude Science bundles data access, compute, and review into one product.
3. Competitive Landscape
| Tool | Type | Note |
|---|---|---|
| Claude Science | Commercial AI workbench (Anthropic) | Managed; ties into Claude model stack |
| openscience | OpenScience | Open-source, model-agnostic workbench |
| Domain models (e.g. alphafold3, isomorphic-labs) | Specialized science models | Narrower, single-task |
Claude Science is the managed/commercial pole versus openscience's self-hosted open-source pole — the same closed-vs-open split playing out across AI-for-science.
4. Unique Observations
- "Claude Code for science" is the strategic move. Anthropic is extending its strongest developer wedge (Claude Code) into the sciences, betting that pharma/biotech R&D budgets are the next large software spend.
- End-to-end judgment as the differentiator. Anthropic's launch example (biotech Manifold) highlighted that Claude Science ranked therapeutic candidates using accumulated program context — not just generating text, but applying domain judgment across a full workflow [1].
- The reviewer system, connectors, and kernels mirror the emerging "agentic science" pattern (see ai-for-science), but delivered as a managed service rather than open infrastructure.
5. Financials / Funding
Product of anthropic — no standalone funding; standalone pricing not disclosed in local research.
6. People & Relationships
- Maker: anthropic (CEO Dario Amodei announced the launch) [3].
- Ecosystem / competitors: openscience (open counterpart), aminer (academic data source), plus domain-model vendors.