Anthropic Launches Claude Science, an AI Workbench for Scientists

Anthropic has released Claude Science in beta, an AI workbench that gathers the tools researchers use daily into one environment. Announced on June 30, 2026, it is available to Claude Pro, Max, Team and Enterprise subscribers on macOS and Linux.

Sixty skills and a coordinating agent

Researchers work with a generalist coordinating agent that has access to more than 60 curated skills and connectors pre-configured for genomics, single-cell analysis, proteomics, structural biology and cheminformatics. These agents can spawn others, including specialist agents built by users, while a separate reviewer agent checks citations and calculations and corrects errors as a pipeline runs.

Scientific data sits across resources such as UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL and GEO, each with its own schema and query language. Claude Science lets a researcher ask in plain language and has specialist agents query and synthesise across those sources. Through NVIDIA's BioNeMo Agent Toolkit it also connects to the BioNeMo models and libraries, including Evo 2, Boltz-2 and OpenFold3.

Reproducible figures, managed compute

The app generates figures and manuscripts beside the code that produced them and renders 3D protein structures, genome browser tracks and chemical structures natively. Every figure comes with its exact code and environment, a plain-language description and the full message history. Edits requested in ordinary language — removing gridlines, switching an axis to log scale — are made by the agent rewriting its own code.

Long jobs such as protein folding are handled the same way: the app drafts a plan, asks before reaching new resources, and submits the work to computing the lab already uses, whether its own HPC cluster over SSH or a Modal account, scaling from one GPU to hundreds. Because the agents run inside a session that keeps context in memory, large datasets are loaded once.

Local data and early results

Claude Science runs on the lab's own infrastructure — a laptop, a Linux machine or an HPC login node — so large or sensitive datasets stay where they are, and only the context needed for each step reaches Claude. A session can be forked at any point to compare two approaches without losing the original thread.

In beta, researchers used it for single-cell RNA sequencing, CRISPR screen design, protein structure prediction and cheminformatics. Stephen Francis of the UCSF Brain Tumor Center said it enabled comprehensive germline workups in roughly one-tenth the previous time, with results his group validated independently. Anthropic is also offering a discounted Team plan for academic and nonprofit labs and will back up to 50 AI for Science projects with up to $30,000 in credits.