Claude Science is Anthropic’s new AI research workbench, launched in beta on June 30, 2026. It connects 60+ scientific databases — spanning genomics, proteomics, structural biology, and cheminformatics — in one environment. It runs on Claude Opus 4.8, costs nothing extra for Pro and above subscribers, and is already in active use at Gladstone Institutes and the Allen Institute.
Anthropic announced Claude Science on Tuesday at an AI for Science briefing — a dedicated workbench that gives scientists one place to do computational research instead of bouncing between dozens of databases, pipelines, and tools. The platform pairs those 60+ pre-connected databases with a multi-agent AI system that works like a research team. The full product details are in Anthropic’s official announcement, first covered by TechCrunch.
One AI That Acts Like a Research Team
The system runs a lead AI as project manager, delegating tasks to specialist sub-agents for different domains — one querying genomics, another analyzing protein structures — or handing off to a custom expert assistant the researcher builds themselves. A separate fact-checker AI reviews every citation and calculation before anything commits to paper. Every figure comes packaged with the exact code, methodology, and message history that generated it — fully auditable and editable in plain English. Sensitive data can also run on the lab’s own infrastructure, never touching Anthropic’s servers.
Already Running in Real Labs
Gladstone Institutes ML scientist Sean Whalen built a genome browser from scratch in days using Claude Science. Allen Institute neuroscientist Jérôme Lecoq used it to build a multi-agent computational review pipeline. Novo Nordisk and the Allen Institute are named official case study partners — a strong signal that major pharma organizations are already running multi-vendor AI strategies.
Anthropic is also offering up to 50 research grants: $30,000 in compute credits per project (plus $2,000 from Modal) for postdoctoral and graduate biomedical research. Applications close July 15, 2026.
The Same Play Anthropic Ran With Claude Code
OpenAI’s GPT-Rosalind, launched in April, is a fine-tuned biological reasoning model gated behind a qualification review for enterprise customers only — partners like Amgen, Moderna, and Novo Nordisk. Google DeepMind owns AlphaFold and AlphaGenome outright, models both rivals can only call as external tools. Anthropic’s answer is to own the operating layer for science the way Claude Code owns software development — and unlike Rosalind, it’s open immediately to all Pro ($17/month annual), Max, Team, and Enterprise subscribers. No waitlist, no approval.
Anthropic’s $35B infrastructure deal with Apollo and Blackstone signals the company is building for vertical product ownership long-term — Claude Science is the first domain-specific workbench in that strategy. It’s available now at claude.ai. For context on the broader Anthropic vs. OpenAI access competition, see how OpenAI’s pricing moves forced Anthropic’s hand, and our coverage of Claude Fable 5 and the platform’s expanding product line.
Last Updated: July 2026
