Claude Science (2026): What It Is, Who Gets It Free, and What Stays on Your Machine

The short answer

Claude Science is Anthropic's workbench app for researchers, in public beta as of September 2026. It is not a new model: it uses the same Claude models your plan already includes, and wraps them in scientific tools, 60+ database connections and the ability to run jobs on your own laptop or cluster. It comes with Claude Pro at $20 a month and up. Verified academic and nonprofit labs can get Team Standard seats for $0.

The part that matters most for anyone with sensitive data is the split in where things live. Your raw datasets and compute stay on your machines. Whatever Claude reads into the conversation goes to Anthropic under standard retention. Everything below is from Anthropic's own pages, read September 25, 2026.

What Claude Science actually is

Anthropic calls it "your research partner for rigorous science." In plain terms, it is an app you install next to your data that can run the analysis, not just talk about it. A general chat assistant can explain single-cell clustering. Claude Science can run it, make the figure, and keep the code that made it.

What ships in the beta:

• Domain specialists for genomics, single-cell, proteomics, structural biology and cheminformatics.
• 60+ scientific databases it can query directly, plus literature search.
• Native viewers for proteins, alignments, genomic tracks, chemical structures and PDFs.
• Provenance on every result. Each figure, table and notebook carries the exact code, environment and conversation that produced it.
• A background reviewer that flags incorrect citations, untraceable numbers, and figures that do not match their underlying code.
• Manuscript drafting beside the analysis, with Markdown and LaTeX previews.

Anthropic also answers the obvious objection itself: it does not replace your specialized tools. Existing Python, R and shell workflows, electronic lab notebooks and internal pipelines plug in as connectors or skills. If you have used Claude artifacts, the provenance idea will feel familiar, except here the artifact is a figure you may have to defend to a reviewer a year from now.

Which plans get Claude Science, and what it costs

The scientists' program is the interesting row. Anthropic says 10,000 scientists can get Claude at no cost to start. A verified principal investigator qualifies for a Team subscription, then adds the group on free Standard seats or $15 Premium seats, for 1 to 25 seats per group. It is limited availability, and final pricing is confirmed only after verification, so treat it as an offer to apply for rather than a price you can budget on. It is also academic and nonprofit only. For the full plan comparison, see my Claude AI pricing breakdown, and Claude Team plan for how the seats work.

How Claude Science actually works: what stays on your machine and what does not

This is the section the launch coverage skips. Claude Science is not a website you upload data to. You install the app wherever the data already lives: your laptop, a lab Linux box, an HPC login node or a cloud VM. You drive it from a browser. It builds the environment each analysis needs, keeps Python and R kernels alive so dataframes stay in memory, and sends heavy jobs to your Slurm cluster over SSH or to your Modal account.

That design is why Anthropic can say "raw datasets and compute stay local." The same sentence has a second half: "content included in prompts and model responses is processed by Anthropic under standard retention." So the rule for a lab with sensitive data is simple. The dataset file stays on your cluster. The rows, summaries and results that Claude reads into the conversation to reason about them do not. If your data agreement cares about that difference, read the retention terms for your plan before the first real dataset goes in.

Two more mechanics worth knowing. It uses your plan's Claude models, so a bigger analysis is a bigger draw on the same usage your plan already meters, covered in my Claude usage limits breakdown. And on Team and Enterprise, nobody gets it until an admin turns it on.

Who should use it, and who should skip it

Use it if you run computational analysis in the life sciences and lose weeks to environment setup, database lookups, and reconstructing which script made which figure. The testimonials on Anthropic's page are all from that world: computational biologists, a UCSF lab that found a virus contaminant in RNA-seq data after a year of searching, drug-discovery teams.

Skip it if your work is not analysis-heavy. For literature questions and writing, regular Claude already does the job on the same plan. And skip the beta for anything regulated until your admin has read the retention terms.

The owner's version

If you run a for-profit R&D team, the free program is not for you, and you will pay normal prices: $25 a Standard seat monthly on Team. A five-person team comes to $125 a month before anyone needs Premium. The app itself costs nothing extra on top of the seat.

The real work is the setup: which plan, who gets which seat, what data is allowed into the conversation, and turning it on for the right people. That is the same decision set as any Claude rollout, and it is what my Claude for Small Business install covers.