Anthropic Launches Claude Science, an AI Workbench Tailored for Researchers

Anthropic has rolled out Claude Science, an AI-powered workbench built to streamline scientific research by pulling together the scattered tools researchers juggle daily. Announced today, the app connects databases like PubMed, UniProt, and the Protein Data Bank (PDB) with compute setups ranging from local laptops to high-performance computing (HPC) clusters, all accessible through a simple chat interface.

Claude Science is available in beta for users on Claude Pro, Max, Team, and Enterprise plans, with academic discounts for nonprofit labs. The company also launched a project support program shelling out up to $30,000 in credits for up to 50 selected AI-for-Science projects.

Consolidating the Fragmented Scientific Workflow

Scientific research typically means hopping between dozens of specialized platforms, each with its own schema, file formats, and data pipelines. Claude Science tackles this by acting as a generalist coordinating agent with access to over 60 curated skills and connectors pre-configured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.

Researchers can query those sources in plain language. Specialist agents then synthesize results across databases such as Ensembl, Reactome, ClinVar, ChEMBL, and GEO, plus preprint servers and domain-specific open models. The platform also connects natively to NVIDIA’s BioNeMo Agent Toolkit, unlocking models including Evo 2, Boltz-2, and OpenFold3.

Reproducible Artifacts with Auditable Histories

Claude Science generates scientific artifacts, figures, manuscripts, 3D protein structures, genome browser tracks, and chemical structures, that are fully reproducible. Each output includes the exact code and environment used to create it, a plain-language description of the methodology, and a full message history. That means researchers can validate and reproduce results months later.

Users can request edits in natural language, like removing gridlines from a figure or flipping an axis to log scale, and the agent dynamically updates its code to match the instruction. A dedicated reviewer agent checks citations, flags untraceable numbers, and verifies that figures match their underlying code, self-correcting as the pipeline chugs along.

Scalable Compute Management

Large analyses, protein folding or genomics pipelines over massive datasets, often force researchers to manually set up compute jobs, monitor them, and retrieve results. Claude Science automates that workflow: it drafts a plan, asks permission before tapping new resources, and submits jobs to the lab’s existing infrastructure, including HPC clusters over SSH or cloud compute via Modal. Scaling from a single GPU to hundreds happens autonomously.

The app keeps a running session that holds context in memory, so hefty datasets only need loading once. It runs on the researcher’s own infrastructure, ensuring sensitive data never leaves the systems they already control. Only the context needed for each step goes to Claude.

“What set Claude Science apart from a general coding assistant was that it could do this end-to-end, gathering the right data and applying the right judgment with the context of past programs built in,” said the team at Manifold Bio, a company designing tissue-targeting medicines that used Claude Science to nominate targets for its latest experiments.

Early Beta Results

Several research groups have already put Claude Science through its paces in beta with notable results. Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent computational review template packing about 20 custom skills. The pipeline reads thousands of papers, extracts central claims and key quantitative findings, and crafts a narrative arc with dedicated sub-agents generating figures and running citation checks. Lecoq’s team previously took up to two years to write such a review; with Claude Science, they have produced roughly 10 reviews, many topping 100 pages, in weeks.

Stephen Francis, an associate professor and epidemiologist at the UCSF Brain Tumor Center, used Claude Science to support studies on the molecular epidemiology of glioma. His lab looks into the genetic basis of germline variants in gliomas. Francis reported that the app enabled comprehensive germline workups across multiple approaches in about one-tenth the previous time, with results independently validated by his group.

Manifold Bio, a biotech firm developing tissue-targeting medicines, used Claude Science to nominate targets for its latest experiments. The platform assessed surface expression, trafficking, and safety across hundreds of targets, ranking candidates against proprietary criteria.

Availability and Project Support

Claude Science is available now in beta on macOS and Linux. Team and Enterprise users need their admin to enable the app. Anthropic is offering discounted seats for active scientific labs at academic institutions and nonprofit research organizations.

The company is also backing up to 50 Claude Science AI for Science projects, each receiving up to $30,000 in credits. Modal will provide up to $2,000 in compute credits for select projects. Applications are open through July 15, 2026, with awards notified by July 31. Projects will run from September 1 to December 1, 2026.

Researchers can get started at claude.com/science or join the AI for Science Discourse community for updates and collaboration.