Anthropic has announced Claude Science, an AI workbench designed specifically for researchers. The application launched in beta on June 30, 2026, and is available to Claude Pro, Max, Team and Enterprise subscribers on macOS and Linux. The product aims to let scientists conduct tasks from literature analysis to large-scale computation in a single environment while producing auditable histories for all outputs.
What Claude Science provides
- It consolidates the disparate tools and file formats researchers commonly navigate, allowing users to work without switching among multiple databases, viewers and custom pipelines. The platform natively renders 3D protein structures, genome browser tracks, chemical structures and other rich scientific artifacts.
- Every generated figure or artifact is accompanied by the exact code and runtime environment that produced it, a plain-language explanation of how it was created, and the full message history, supporting later validation and reproducibility.
- Users can request edits to figures or code in natural language (for example, removing gridlines or switching an axis to a log scale), and the agent will update its own code accordingly.
Agents, skills and connectors
Users interact with a generalist coordinating agent that has access to more than 60 curated skills and connectors preconfigured for genomics, single-cell, proteomics, structural biology, cheminformatics and other life‑science domains. These agents can spawn sub‑agents and incorporate specialist agents created by users; a separate reviewer agent inspects citations and computations, flagging and correcting errors.
Compute management and data security
Claude Science manages computational jobs: it drafts a plan, asks for permission before accessing new resources, and submits jobs to the lab’s existing compute infrastructure (for example, an HPC cluster over SSH or a Modal account for on‑demand compute). The system is scalable, from a single GPU up to hundreds of GPUs as needed.
Because agents run inside an active session that retains context in memory, large datasets need to be loaded only once. Claude Science runs on the lab’s infrastructure—laptop, Linux box, or HPC login node—so large or sensitive datasets do not have to leave their systems; only the contextual information required for each analysis step is sent to Claude. As pipelines run, a reviewer agent inspects outputs, identifying untraceable numbers, incorrect citations and figures that do not match their underlying code. Users can fork a session at any point to compare approaches without losing the original thread.
Domain‑ready configuration and data sources
Claude Science is preconfigured for genomics, single‑cell, proteomics and cheminformatics, and connects to more than 60 scientific databases. Specialist agents query and synthesize across sources such as UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL and GEO, as well as journals, preprint servers and domain‑specific open models. Claude Science uses NVIDIA’s BioNeMo Agent Toolkit to connect natively to BioNeMo models and libraries, including Evo 2, Boltz‑2 and OpenFold3.
Researchers’ existing models, datasets and pipelines can also be connected: any pipeline can be saved as a reusable skill, and preferred lab tools can be accessed via connectors so future sessions inherit those resources automatically.
Early user examples from the beta
-
Manifold Bio: Manifold Bio, which designs tissue‑targeting medicines, used Claude Science to nominate targets for experiments. For each tissue and target the system evaluated surface expression, trafficking and safety, and ranked candidates against Manifold’s proprietary internal data. Manifold highlighted that Claude Science was able to run the end‑to‑end workflow, gathering the right data and applying context from past programs.
-
Jérôme Lecoq (Allen Institute): Lecoq and colleagues built a multi‑agent “computational review template” of roughly 20 custom skills for writing long‑form reviews. Sub‑agents read thousands of papers, extract central claims and key quantitative findings into an evidence state database, and then the pipeline composes the review section by section, delegating each to specialized sub‑agents. Within sections dedicated agents generate cross‑study quantitative figures directly from the evidence database. The workflow leverages actor‑critic pairs—one agent creates content while a separate reviewer agent evaluates accuracy and citation fidelity. Lecoq’s team previously took up to two years for such reviews; using Claude Science they have produced about ten reviews, many over 100 pages, with citations checked by reviewer agents.
-
Stephen Francis (UCSF Brain Tumor Center): Stephen Francis’s group studies the molecular epidemiology of glioma and how many small‑effect germline variants together influence individual susceptibility. Francis said that while the work predated Claude Science, the app has dramatically accelerated analyses: comprehensive germline workups across multiple approaches now take roughly one‑tenth of the previous time. His group independently validated Claude Science’s results.
Access, grants and support
The Claude Science app is available in beta on macOS and Linux for Claude Pro, Max, Team and Enterprise plans; Team and Enterprise administrators must enable the app for their organizations. Anthropic offers discounted Team seats for active scientific labs at academic institutions and nonprofit research organizations.
Anthropic is supporting up to 50 “Claude Science AI for Science” projects, providing up to $30,000 in credits per project, and Modal will provide up to $2,000 in compute for selected projects. Applications are open through July 15, 2026, with award notifications by July 31, 2026. Supported projects will run from September 1 to December 1, 2026.
To follow product announcements, give feedback and join the community, users are invited to the AI for Science Discourse community. Further information and getting‑started resources are available at claude.com/science.
