Claude Science

Claude Science is Anthropic's research app for scientists, a public beta since Jun 30, 2026: it runs analyses on your own machines or cluster, shows the code behind every figure, renders proteins and genomes, and queries more than 60 scientific databases. It comes with Pro, Max, Ultra, Team and Enterprise, and academic labs can get a discounted Team plan.

Claude Science is a desktop and browser app for research, not a model: it runs on the same Claude models your plan includes and works where your data is, on a laptop, a lab server, a cluster login node or a cloud machine (Claude Science, read Nov 6, 2026). Anthropic introduced it on Jun 30, 2026 as "an AI workbench for scientists", and it is a public beta on the Pro, Max, Ultra, Team and Enterprise plans. Its distinctive idea is provenance: every figure, table and notebook carries the code, environment and conversation that produced it, and a background reviewer flags citations and numbers it cannot trace. Scientists at academic and nonprofit institutions can get a discounted Team plan that includes it. It replaces neither your specialized tools nor your judgment; Anthropic positions it as a layer that drives those tools.

Quick facts

What it is	A public beta app, not a model
Introduced	Jun 30, 2026
Plans	Pro, Max, Ultra, Team, Enterprise; not Free (as of Nov 6, 2026)
Platforms	macOS (Apple Silicon and Intel), Windows, Linux
Databases	More than 60 scientific databases
Skills and connectors	More than 60 preconfigured for genomics, proteomics, cheminformatics and more
Compute	Your laptop, your HPC cluster over SSH, or your Modal account
Discounted plan	Claude Team plan for scientists

What Claude Science does

You describe an analysis; Claude Science drafts a plan, asks before using new compute, writes and runs the code, and returns results as scientific artifacts you can inspect (Introducing Claude Science, Jun 30, 2026). A coordinating agent can start other agents and work with specialist agents you create. Python and R kernels persist through a session, so data frames and loaded models stay in memory, and a session can be forked to compare two approaches (Claude Science, read Nov 6, 2026).

The Claude Science beta app with a sidebar of sessions such as Samosa Manuscript, Samosa Editorial Review and Samosa Model Training. The middle pane shows samosa_fig1a.png, a scatter plot of 138 species and 5,672 cell types colored as neuron, muscle and immune, with a comment box reading these labels are hard to see and a Send button. The right pane, version 2, has tabs Code, Execution Log, Messages, Environment and Review, a Download script button, input CSV files and the matplotlib code. — Claude Science shows a cross-species cell atlas figure with a reader's comment on it, next to the Python code that drew it. Anthropic

0:54 Anthropic lists workflows such as single-cell RNA-seq, phylogenetics, protein structure and protein language models, and cheminformatics and molecular design. Manuscripts can be drafted next to the analysis with Markdown and LaTeX previews. Pipelines you build can be saved as Skills that later sessions pick up.

Provenance and the reviewer

0:14 Every figure, table and notebook keeps the code, the environment, a plain-language description and the full conversation behind it (Claude Science, read Nov 6, 2026). A reviewer agent runs in the background and flags wrong citations, numbers it cannot trace to code, and figures that do not match their code. You can annotate a figure in plain language ("make the axis log scale"), and Claude edits the code that produced it rather than the image.

The Claude Science beta app in a Literature review session. The user asks for a review of cross-species single-cell RNA-seq integration as a LaTeX document and compiled PDF. Claude dispatches five parallel retrieval tracks (PubMed, bioRxiv, OpenAlex, CELLxGENE and orthology-free embedding methods); a Reviewer card shows a Warn finding that one PubMed ID was assigned to two methods, which Claude acknowledges and corrects. On the right, review.pdf shows the paper's title page with a citation chart and two-column text. — Claude Science writes a literature review as a compiled PDF while a reviewer agent flags a mismatched citation in the plan. Anthropic

These checks reduce, but do not remove, the usual risks of an AI assistant in research: a reviewer agent built on the same model family can miss the same errors, so results still need a scientist's check. Anthropic's case studies, such as UCSF's report of germline analyses in about one-tenth of the time, are the participants' own reports; that group says it validated its results independently.

Renderers

Claude Science shows proteins and 3D structures, sequence alignments, genome browser tracks, chemical structures and PDFs natively, with nothing extra to install (Claude Science, read Nov 6, 2026).

Integrations

It queries more than 60 scientific databases, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL and GEO, along with journals and preprint servers (Introducing Claude Science, Jun 30, 2026). Named tools include Benchling, BioRender, bioRxiv, Consensus and Clinical Trials, and LatchBio and Helix connect through MCP. Skills from NVIDIA's BioNeMo Agent Toolkit reach the models Evo 2, Boltz-2 and OpenFold3. Your own internal APIs, electronic lab notebooks and pipelines connect as Connectors or skills (Claude Science).

Compute and data handling

0:39 The app runs on infrastructure you choose. It writes batch scripts and submits jobs over SSH to your own machine or HPC cluster (Slurm), or to your Modal account, from one GPU to hundreds (Introducing Claude Science). Raw datasets and compute stay on your machines or cloud accounts, and only the context each step needs is sent to Claude; prompts and responses are processed by Anthropic under its standard retention (Claude Science). See Claude Privacy and Data Use for those terms.

Plans and platforms

The June announcement listed Pro, Max, Team and Enterprise, on macOS or Linux (Introducing Claude Science, Jun 30, 2026). The product page, read Nov 6, 2026, adds Ultra (split from Max in late Oct 2026) and Windows (Claude Science). On Team and Enterprise an admin must turn it on first; Enterprise adds SSO, SCIM, custom roles, usage analytics and the Compliance API. It is not on the Free plan. Prices are on Claude Plans and Pricing.

Some sensitive life-sciences requests may be handled by a different Claude model, as Anthropic's safeguards do elsewhere, and vetted research organizations can apply for wider access through Anthropic's trusted-access programs (Claude Science), such as the Life Sciences Verification Program behind Claude Mythos.

The discounted Team plan for scientists

Principal investigators at academic and nonprofit research institutions can get a Claude Team plan for their group with Claude Science on by default (Claude Team plan for scientists, read Nov 6, 2026). As of Nov 6, 2026 the promotional prices, for 12 months, are:

Seat	Promotional price	Regular price
Standard	$0 a month	$20 a month
Premium (5x the usage)	$15 a month	$100 a month

Source: Claude Team plan for scientists, read Nov 6, 2026.

A group can have 1 to 25 seats, availability is limited, and the final price is confirmed after verification. For-profit companies, contract research organizations and industry R&D teams are pointed to the regular Team or Enterprise plans. Larger projects can apply for extra credits through Anthropic's AI for Science program; its 2026 round offered up to $30,000 in credits to up to 50 projects, with applications closed on Jul 15, 2026.

Frequently asked questions

Is Claude Science a model? No. Claude Science is a public beta app that runs on the same Claude models your plan includes, with the same safeguards.

Is Claude Science free? Claude Science needs a Pro, Max, Ultra, Team or Enterprise plan; academic and nonprofit groups can get it through the Claude Team plan for scientists, whose Standard seats cost $0 a month for the first 12 months as of Nov 6, 2026.

Does Claude Science upload my data to Anthropic? Claude Science keeps raw datasets and compute on the machines or cloud accounts you choose and sends Claude only the context each step needs; prompts and responses are processed by Anthropic under standard retention.

Does Claude Science run on Windows? Yes. As of Nov 6, 2026 Claude Science is available for macOS, Windows and Linux, and can also be installed on a cluster login node or cloud machine and used from a browser.

Is Claude Science the same as Claude for Life Sciences? No. Claude for Life Sciences was an Anthropic launch for researchers on Oct 20, 2025; Claude Science is a separate app, introduced on Jun 30, 2026, that runs analyses and tracks their provenance.

Sources

Introducing Claude Science, an AI workbench for scientists, Anthropic, Jun 30, 2026
Claude Science, claude.com, read Nov 6, 2026
Claude Team plan for scientists, claude.com, read Nov 6, 2026
Introducing Claude Science (now in beta), Claude, Jun 30, 2026 (video)
Claude for Life Sciences, Anthropic, Oct 20, 2025
Plans and pricing, claude.com, read Nov 6, 2026

Moments

Introducing Claude Science (now in beta)

Claude

0:54 Anthropic lists workflows such as single-cell RNA-seq, phylogenetics, protein structure and protein language models…
0:14 Every figure, table and notebook keeps the code, the environment, a plain-language description and the full…
0:39 The app runs on infrastructure you choose. It writes batch scripts and submits jobs over SSH to your own machine or…