Anthropic released Claude Science in beta on June 30, 2026, giving researchers a single AI workbench that unifies literature analysis, data pipeline execution, figure generation, and manuscript drafting. The app is available now for Claude Pro, Max, Team, and Enterprise users on macOS and Linux, and it arrives as labs face growing pressure to produce reproducible, auditable research across dozens of fragmented tools.
What the platform does
Claude Science puts a generalist coordinating agent in front of scientists, with access to more than 60 curated skills and connectors pre-configured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. The agent spins up sub-agents for specialized tasks and a reviewer agent checks citations and calculations, flagging and correcting errors as they appear. Every figure, manuscript, and analysis carries a plain-language description of how it was created, the exact code and environment that produced it, and the full message history.
Unlike a general coding assistant, Claude Science natively renders scientific artifacts-3D protein structures, genome browser tracks, chemical structures-and lets researchers edit figures by asking in plain language. "Remove the gridlines" or "switch the y-axis to log scale" triggers the agent to edit its own code and regenerate the output.
Compute and data handling
Large analyses often force researchers to stop their work, configure cluster jobs, wait for results, and retrieve outputs. Claude Science drafts a plan, asks before reaching additional resources, and submits jobs to the computing infrastructure a lab already uses-an HPC cluster over SSH or a Modal account for on-demand compute. It scales from a single GPU to hundreds as needed, and because agents run inside a session that holds context in memory, large datasets load once.
Data stays on the lab's own systems. Only the context needed for each step of an analysis is sent to Claude, which matters for labs handling sensitive patient data or proprietary datasets. The reviewer agent inspects pipeline outputs in real time, flagging incorrect citations, untraceable numbers, and figures that don't match their underlying code.
Domain-ready integrations
Scientific data is scattered across sources like UniProt, PDB, Ensembl, ClinVar, ChEMBL, and GEO, each with its own schema and query language. When a researcher asks a question in plain language, specialist agents query and synthesize across all of these sources. The platform uses skills from NVIDIA's BioNeMo Agent Toolkit to connect natively to life sciences models including Evo 2, Boltz-2, and OpenFold3.
Labs can also connect their own models, datasets, and pipelines, saving any pipeline as a reusable skill or connector that future sessions inherit. This means researchers access Claude, proprietary data, and validated in-house tools in a single conversation. Scientists who want structured guidance on integrating AI into their workflows can explore AI for Science & Research Courses alongside platform adoption.
What early adopters are building
Manifold Bio, a company designing tissue-targeting medicines, used Claude Science to nominate targets for its latest experiments. The platform assessed surface expression, trafficking, and safety for each tissue and target, ranking candidates against criteria drawn from Manifold's proprietary data. What set it apart, the company said, was the ability to do this end-to-end-gathering data and applying judgment with context from past programs.
Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent system with roughly 20 custom skills to produce long-form scientific reviews. Sub-agents read thousands of papers, extracted central claims and quantitative findings, and stored them in an evidence database. A narrative arc is then constructed, with dedicated agents writing each section and generating cross-study figures directly from the evidence. Actor-critic pairs-one agent creates, a reviewer agent evaluates for accuracy and citation fidelity-form the backbone. Before Claude Science, reviews like this could take two years. Lecoq now has about 10 reviews, many exceeding 100 pages, with citations checked by reviewer agents.
Stephen Francis, an associate professor and epidemiologist at the UCSF Brain Tumor Center, applied the platform to glioma molecular epidemiology. His lab investigates how thousands of small-effect germline variants combine to shape tumor susceptibility. Francis said the app cut analysis time to roughly one-tenth of what it previously took, and his group independently validated Claude Science's results, confirming the analyses were both fast and accurate.
Beta access and research grants
Claude Science is available on macOS and Linux for Pro, Max, Team, and Enterprise plans. Team and Enterprise users need admin approval. A discounted Team plan is now offered for active academic and nonprofit research labs. Anthropic is also supporting up to 50 AI for Science projects with credits up to $30,000, and Modal is providing up to $2,000 in compute for select projects. Applications close July 15, 2026, with award notifications by July 31. Projects run September 1 through December 1, 2026.
Why this matters for researchers
Claude Science targets the friction that eats into research time: switching between PubMed, Jupyter, R, cluster terminals, and specialized databases, then manually tracking how each result was produced. By collapsing these into a single environment with auditable provenance-and keeping data on infrastructure labs already control-the platform removes steps that don't advance the science. The early adopter results suggest the time savings are real and measurable, not aspirational. Researchers who want to evaluate whether this fits their workflow can start with the beta now, while the grant program offers funded runway for teams pushing into multi-agent methods and large-scale analysis.
Your membership also unlocks:
