What Is Claude Science? Anthropic's AI Research Workbench 2026
What Is Claude Science?
Claude Science is an AI research workbench launched by Anthropic on June 30, 2026, designed to accelerate scientific discovery. Rather than releasing another model, Anthropic built an integrated environment that consolidates fragmented scientific tools — literature search, data analysis, visualization, manuscript drafting, and research computing — into a single, unified interface.
Claude Science was announced alongside Claude Sonnet 5 (GA on June 30, 2026) and the reopening of Claude Fable 5 access (restored July 1, 2026). MIT Technology Review called it “Anthropic’s newest flagship product,” and TechCrunch noted it “bets on workflow, not a new model, to win over scientists.”
Key Features
Multi-Agent Research Architecture
Claude Science uses a coordinating AI agent with access to over 60 scientific skills and connectors, covering genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. It can fetch primary literature from PubMed and other databases, then design and execute follow-up experiments autonomously.
Native Scientific Displays
Unlike general-purpose chatbots, Claude Science can natively render:
- 3D protein structures (PDB files rendered interactively)
- Genome browser tracks (BAM/CRAM coverage, peak calls)
- Chemical structures (SMILES, SDF rendering)
- Heatmaps, volcano plots, and PCA embeddings
Code-Enabled Reproducibility
Every figure, table, and notebook includes the source code, computing environment, plain-language process description, and full conversation history. Researchers can request changes in plain English, and Claude Science automatically edits the underlying Python or R code. This addresses one of the biggest pain points in computational research: irreproducible figures.
Reviewer Agent
A built-in reviewer agent systematically checks:
- Citation accuracy — does the cited paper actually support the claim?
- Numerical consistency — do the numbers in the text match the code output?
- Figure-code alignment — could this figure have been produced by the attached code? Results lacking supporting evidence are flagged for human review.
Compute Infrastructure
Claude Science can execute workflows on:
- Local laptops
- Laboratory Linux systems
- Cloud virtual machines
- HPC login nodes
Applications for the AI for Science grant program (up to $30,000 in Claude credits, plus $2,000 in Modal compute) close on July 15, 2026.
Pricing and Availability
Claude Science does not require a separate subscription. It is available to all paid Claude users:
| Plan | Monthly Price | Claude Science Access |
|---|---|---|
| Free | $0 | Not available |
| Pro | $20 | Full access |
| Max 5x | $100 | Full access (50% Fable 5 usage through July 7) |
| Max 20x | $200 | Full access |
| Team | $25/seat/mo | Full access (admin must enable) |
| Enterprise | Custom | Full access (admin must enable) |
As of July 7, 2026, Claude Science uses underlying Claude models at their standard API rates, with usage counting toward plan quotas.
Claude Science vs Traditional Research Tools
Claude Science’s advantage over tools like ChatGPT Deep Research, Perplexity, and Google’s NotebookLM is its execution capability. Where other research tools mostly read and summarize, Claude Science can write code, execute it, analyze results, and synthesize all of it into a publishable figure or manuscript section — all within the same session.
Early benchmarks (July 1–7, 2026) show promising results on GeneBench-Pro, a new benchmark measuring agentic computational biology performance. GPT-5.6 Sol leads at ~29–32% pass rate on tasks that typically cost thousands of dollars in human expert time.
Limitations
- Not a standalone API — requires a Claude subscription
- Computational biology focus — strongest in bioinformatics, genomics, and cheminformatics; weaker in experimental planning
- Grant deadline — AI for Science project applications close July 15, 2026
- Still in beta — Anthropic describes it as a “public beta” as of July 1, 2026