Scientific research has long been slowed by fragmented tools and manual handoffs. Researchers navigate databases with different schemas, switch between Jupyter, R, and cluster terminals, and spend hours setting up compute jobs. Anthropic’s newly launched Claude Science directly addresses this pain by offering an integrated AI workbench that streamlines every stage of discovery.
Released today in beta, Claude Science is not just another coding assistant. It bundles over 60 pre-configured skills for genomics, proteomics, structural biology, and cheminformatics into a single environment. A coordinating agent delegates tasks to specialized agents, while a separate reviewer agent checks citations and calculations in real time. Every output comes with an auditable history of its creation, so results can be validated and reproduced months later.
One of its strongest features is flexible compute management. Scientists can run large analyses—like protein folding or genomics pipelines—without manually setting up jobs. The agent drafts a plan, requests permission before accessing resources, and submits tasks to existing HPC clusters or cloud services like Modal. Data stays on the lab’s own infrastructure, addressing security and privacy concerns for sensitive datasets.
Domain readiness is another differentiator. Claude Science connects natively to NVIDIA’s BioNeMo Agent Toolkit, granting access to models such as Evo 2, Boltz-2, and OpenFold3. It also allows integration with proprietary tools and pipelines, saving them as reusable skills for future sessions. This bridges the gap between general-purpose AI and specialized scientific needs.
Early adopters report dramatic gains. Manifold Bio, a biotech firm designing tissue-targeting therapies, used Claude Science to nominate targets for experiments by evaluating surface expression, trafficking, and safety across hundreds of candidates. Unlike generic coding aids, Claude Science performed the full end-to-end analysis while incorporating context from prior programs.
At the Allen Institute, neuroscientist Jérôme Lecoq built a multi-agent system that reads thousands of papers, extracts central claims, and writes review articles section by section. An actor-critic pair ensures accuracy, with one agent creating content and another verifying citations. A process that once took two years can now be completed in months. Lecoq’s team has produced over ten reviews, each exceeding 100 pages, with citations rigorously cross-checked.
Stephen Francis, an epidemiologist at UCSF, applied Claude Science to molecular epidemiology of glioma. His lab studies how small-effect germline variants combine to influence cancer susceptibility. Claude Science cut analysis time by roughly 90% while maintaining robustness, with independent validation confirming the results.
The integration of AI into science is not just about speed—it’s about enabling questions that were previously impossible to ask. By automating tedious steps and ensuring reproducibility, Claude Science gives researchers more time for creative hypothesis generation. Some critics worry about over-reliance on AI, but the built-in reviewer agent and full audit trail offer a measured path forward.
For scientists interested in adopting the platform, Claude Science beta is available on macOS and Linux for Pro, Max, Team, and Enterprise users. Anthropic is also funding up to 50 projects with $30,000 in credits, plus additional compute from Modal. Applications are open through July 2026, targeting biology and biomedical research initially.
As feedback rolls in, Claude Science will continue to evolve. It may soon become a standard tool in labs worldwide, transforming how scientists interact with data, literature, and computation. The era of AI-assisted discovery is already here, and it runs on your own infrastructure.