Anthropic Launches Claude Science Beta: An AI Workbench Built to Compress Years of Research Into Days

Last fall, Anthropic announced a push into life sciences. Six months later, Claude Science is the result — not a new model, but a research environment designed to eliminate the context-switching and pipeline friction that consume a scientist's day.

The Problem: Research Tools Are Fragmented

A typical day for a computational biologist involves shuttling between PubMed, Jupyter, R, an HPC terminal, and a roster of domain databases — each with its own schema and query language. Submitting a cluster job, waiting for results, and pulling them back often takes longer than the analysis itself. Claude Science collapses these steps into a single conversation, from literature review through multi-step execution to figure generation and manuscript refinement. Every output carries a full reproducibility trail: the exact code, environment, and message history that produced it.

How It Works: Three-Tier Agent Architecture

Claude Science runs three tiers of agents working in concert:

• Coordinating Agent: Interprets a plain-language research question, devises a strategy, and delegates to the right specialist agent from a library of 60-plus curated skills and connectors.
• Specialist Agents: Query and synthesize across domain-specific sources — UniProt, PDB, Ensembl, ClinVar, ChEMBL, GEO, and more — and tap into NVIDIA's BioNeMo Agent Toolkit for access to life sciences models including Evo 2, Boltz-2, and OpenFold3.
• Reviewer Agent: Runs continuously, flagging incorrect citations, untraceable numbers, and figures that don't match their underlying code — and self-correcting as the pipeline runs.

Researchers can fork a session at any point to compare two approaches without losing the original thread. Because agents hold context in memory across a running session, even massive datasets only need to be loaded once.

Early Results: Cutting Years Down to Weeks

Anthropic shared three concrete cases from the beta period:

Manifold Bio (tissue-targeting medicines): The company used Claude Science to nominate targets for its latest experiments, assessing surface expression, trafficking, and safety for hundreds of targets — end-to-end, incorporating proprietary internal data as context — in a workflow that previously required manual coordination across multiple tools.

Jérôme Lecoq, Allen Institute (neuroscientist): Built a multi-agent "computational review template" with about 20 custom skills. Sub-agents read thousands of papers, extracted central claims and key quantitative findings into an evidence database, then constructed a narrative review section by section — each section written by a specialized sub-agent, with actor-critic pairs checking accuracy and citation fidelity. Reviews that once took his team up to two years are now produced in far less time, many running over 100 pages.

Stephen Francis, UCSF Brain Tumor Center (epidemiologist): Used Claude Science to accelerate molecular epidemiology studies of glioma. Comprehensive germline analyses across multiple approaches now complete in roughly one-tenth the previous time. His group independently validated the results.

What's Behind It: A $400M Acquisition and a Nobel Laureate

Two strategic moves set the stage for Claude Science.

In April, Anthropic acquired Coefficient Bio for approximately $400 million in stock. The stealth startup — fewer than ten employees, mostly former Genentech computational biologists — brought deep expertise in protein design and biomolecule modeling.

On June 19, John Jumper announced he was leaving Google DeepMind for Anthropic. Jumper shared the 2024 Nobel Prize in Chemistry with DeepMind's Demis Hassabis for co-developing AlphaFold, the protein structure prediction system. Recruiting one of the field's most decorated scientists signals that Anthropic's science ambitions extend well beyond tooling.

Availability and Grants

Claude Science beta is available at claude.com/science for all Pro, Max, Team, and Enterprise plan users on macOS and Linux. Team and Enterprise admins must enable it first. Academic and nonprofit research labs can apply for discounted Team plan seats.

AI for Science Grants: Anthropic is funding up to 50 projects with grants of up to $30,000 in Claude Science credits. Modal is also providing up to $2,000 in compute for select projects. Applications are open through July 15, 2026, with award notifications by July 31. Projects run September 1 – December 1, 2026. The early focus is on biology and biomedical research — apply at the link below.

Why It Matters

Claude Science is Anthropic's attempt to occupy the operating layer for scientific research the way Claude Code has for software development. Rather than selling model access, it's building the environment researchers work in — which makes switching costs higher and domain-specific feedback loops tighter.

The harder question is whether the 10x productivity claims will hold across diverse research contexts beyond the beta cohort. Independent replication studies and peer-reviewed benchmarks will be the real test. For now, the early signals — a Nobel laureate hire, a $400M acquisition, and credible early-adopter case studies — suggest Anthropic is serious about this space.