← Back to feed

Anthropic launched Claude Science on June 30, a research workbench that puts a coordinating AI agent, 60 scientific databases, HPC compute submission, and molecular rendering inside one persistent session. The product is in beta and is Anthropic’s first dedicated application for the life sciences vertical — an initiative the company has been developing since October 2025.

What the Product Does

The core problem Claude Science targets is context fragmentation. Computational biologists typically work across databases, notebooks, R environments, cluster queues, and visualisation tools simultaneously. Every switch breaks state and forces manual rechecking. Claude Science consolidates that loop.

A top-level coordinating agent dispatches specialist sub-agents for individual tasks — literature retrieval, dataset querying, sequence analysis, figure generation — and stitches results back into a single session. The app renders 3D protein structures, genome tracks, chemical structures, and manuscript drafts inside the same view as underlying code.

Key capabilities:

• 60 scientific databases accessible without leaving the session
• HPC and Modal compute integration: submit jobs from 1 GPU to hundreds; datasets stay local
• Code-traced artifacts: every result includes its source code, execution environment, plain-language method description, and the full message history that produced it
• Reviewer agent: automatically checks calculations, references, and figures against their originating source code before the user sees them

Reproducibility is the explicit design target. By keeping code, environment, and output in the same artifact, Claude Science makes verification a structural property rather than a user responsibility.

Early Performance

Early users from pharma research reported producing 10 review drafts exceeding 100 pages each and completing germline analyses in one-tenth the time compared to their previous workflows. Anthropic has not published a formal benchmark evaluation; these figures come from beta user disclosures.

TechCrunch described the product as betting on “workflow, not a new model, to win over scientists” — the underlying model is Claude’s existing frontier stack, not a biology-specific model trained from scratch.

The Strategic Frame

Claude Science is Anthropic’s second major vertical product after the Glasswing cybersecurity initiative. Life sciences fits the Anthropic playbook: high-value professional users, long sessions, complex multi-step reasoning, and output that requires auditability.

The IPO-bound company is building out enterprise verticals ahead of its October Nasdaq listing. At $16.20 per user per month in average revenue — compared to $2.20 for OpenAI’s user base — Anthropic’s enterprise focus is already showing in revenue per seat. A defensible position in pharma and biotech research, where switching costs are high and compliance requirements favour verifiable output, extends that advantage.

Anthropic did not disclose pricing for Claude Science. The product is currently available in beta to selected research institutions and pharma partners. Broader access timeline was not announced.