What is Claude Science and what does it cost?

What is Claude Science?

Claude Science is a desktop app in beta that pairs Claude with an analysis environment on your computer. It is an app, and Anthropic's product page says it is not a model. It is included in the Pro, Max, Ultra, Team and Enterprise plans with nothing separate to buy, and it is not available on the Free plan.

You describe a research task in plain language. Claude writes and runs Python, R or shell code in a sandbox, which is a restricted space that keeps the code away from the rest of your system. Claude reads the folders you grant it and pulls data from scientific databases through connectors. It saves results as versioned artifacts with a provenance record, which is a log of how each result was made. A background reviewer can then check Claude's claims against the work that actually ran.

Your files stay on your computer by default. You approve each new folder, network host and remote job before Claude can use it. You install Claude Science separately from the Claude desktop app. It is also wider than biology: the Claude Science overview lists chemistry, physics, engineering, mathematics, machine learning and the social sciences, among other fields.

The facts in this guide come from Anthropic's docs and help pages as read on 5 November 2026. Claude Science is in beta and changes often. The latest changelog entry at that date was version 0.1.71, from 27 October 2026.

What does Claude Science cost?

Claude Science has no price of its own. You pay for a Claude plan, and the app comes with it. If you do not have a plan yet, read what every Claude plan costs first, then which Claude plan you need.

The real cost is usage. Your Claude Science usage counts toward the same limits as the rest of your plan, including Claude Code and Cowork. This means a long analysis in Claude Science leaves less of the limit for your other Claude work. In the app, Settings > Usage shows how much you have used and when each limit resets.

The overview page does not give per-plan numbers for Claude Science, such as messages or tokens per week. Use Settings > Usage for your own figures, and read how Claude usage limits work for the general rules.

When do you pay more than the plan price?

The docs name only these cases:

• Usage credits and extra usage. On Pro, Max and Ultra you can turn on usage credits in your claude.ai settings, so you can work past your limits. You can then turn them off or change your monthly spend limit under Settings > Usage in Claude Science. Team plans call the same option extra usage, and your admin turns it on. On Team and Enterprise, your admin sets the spend limits.
• Usage-based Enterprise plans. These plans have no fixed usage limits. Your organization pays for usage on top of the seat price, within the spend limits your admin sets.
• Services you connect with your own account. Modal, a cloud GPU (graphics processing unit) service, is the example in the docs. Modal bills you directly, and Anthropic does not bill for it.

One more detail affects the bill. Some models need usage credits on some plans, and Claude pricing lists which models each plan includes.

What happens at the 5-hour and weekly limits?

Plans with usage limits have two: a limit that resets every 5 hours, and a weekly limit. Anthropic's usage limits article says the 5-hour period starts when you send your first message, and your limit resets in full five hours later. Verified Ultra accounts do not have a 5-hour limit.

As you approach a limit, Claude Science asks whether the session should continue. At the 5-hour limit you can choose Keep going or hold the session until the reset. At the weekly limit, the warning offers only Keep going.

What happens at the limit depends on the reset time and on your credits:

• No credits, reset less than 8 hours away. The session pauses, and the composer shows when it will resume. The session continues by itself after the reset, as long as Claude Science stays open.
• No credits, reset more than 8 hours away. The session pauses for up to 8 hours, then stops and shows when the limit resets. This can happen at a weekly limit. After the reset, select Resume or send a message.
• Usage credits or extra usage turned on. By default, Claude Science asks whether to keep going on them, and the session waits until you answer. If you say no, the session waits for the reset, or stops if the reset is more than 8 hours away.

Your conversation and artifacts are kept in each of these cases. For a long overnight run, the practical point is this: keep the app open, because the automatic resume depends on it. If you turned credits on, also expect the session to wait for your answer at the limit. For options outside the app, see what to do when you reach a Claude limit.

Claude Science vs Claude Code vs Cowork: which job does each one do?

All three use your plan's limits, so the choice is about the job.

• Claude Code is for software work. Anthropic describes it as an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools. It runs in the terminal, an IDE (integrated development environment), the desktop app and the browser. See what Claude Code is.
• Cowork is for multi-step tasks done for you, with no terminal. It uses the same agentic architecture as Claude Code, and it is on the paid plans. See which plans include Cowork.
• Claude Science is for research analysis. It runs Python, R or shell code in a sandbox on your computer. It adds scientific connectors, a provenance record and the reviewer.

A simple way to choose is to look at the result you want. If the result is working software, use Claude Code. If the result is a finished task and you do not want a terminal, use Cowork. If the result is a figure or a finding that you must be able to trace back to the code that made it, use Claude Science.

Does Claude Science run on Windows, macOS and Linux?

Yes. The current docs list macOS 13 or later (Apple silicon or Intel), Windows 11 (x64) and Linux x64 on a glibc-based distribution. You also need about 5 GB of free disk space for the runtime and starter environments, and you must be in a country or region where Anthropic offers claude.ai.

Be careful with older sources. Anthropic's launch post from 30 June 2026 says macOS and Linux, and its plan list has no Ultra. Those lines are now out of date. The changelog shows Windows support arrived in version 0.1.47 on 10 September 2026, and the current docs list both Windows 11 and Ultra.

On Windows, the docs also describe running the Linux command-line version under WSL 2 (Windows Subsystem for Linux) with Ubuntu 24.04 or later. On Linux, do not expect a normal app window. You start it from the terminal with claude-science serve, and it opens in a browser tab.

What does Claude Science need on Linux?

Claude runs code inside a sandbox, and the sandbox depends on system parts that a normal Linux install may not have. The docs list these requirements:

• An x64 machine with a glibc-based distribution. glibc is the GNU C library. The arm64 architecture and musl-based distributions such as Alpine are not supported.
• socat, a small tool that relays network connections.
• bubblewrap 0.8.0 or later, the tool that creates the sandbox.
• Unprivileged user namespaces allowed by the kernel. This kernel feature lets a normal user create an isolated environment without root.

On Ubuntu or Debian, install the packages with the command from the docs. This step needs sudo.

sudo apt-get update && sudo apt-get install -y curl bubblewrap socat
/usr/bin/bwrap --version

The second command should print version 0.8.0 or later. Ubuntu 24.04 ships a new enough version, and Ubuntu 22.04 does not. The changelog for 0.1.71 says Ubuntu 22.04 and earlier, Debian 11 and earlier, RHEL (Red Hat Enterprise Linux) and Rocky 8 and 9, and Amazon Linux 2 ship an older bubblewrap or none. If Claude Science cannot set up the sandbox, it refuses to start, because it will not run code unsandboxed.

Installing the app itself does not need root:

curl -fsSL https://claude.ai/install-claude-science.sh | bash
. ~/.profile
claude-science --version

The installer verifies a checksum and puts the claude-science command in ~/.local/bin. The last command should print a version number. command not found: claude-science means that folder is not on your PATH yet.

Two failures are worth knowing before you start:

• cannot create unprivileged user namespaces. The kernel or an AppArmor profile blocks the sandbox. The docs say some Ubuntu 24.04 images restrict this, so a new enough bubblewrap does not guarantee a working sandbox. The error message names the settings to check.
• Package downloads are being blocked by network policy. This can appear even when your network blocks nothing. In version 0.1.71 and later, Claude Science runs bubblewrap and socat only from /usr/bin, /usr/local/bin or /bin, or from a copy you approved. It does not search your PATH. A socat outside those system folders leaves the sandbox with no network, so the starter environments cannot download their packages. Install socat as a system package to fix this.

If you have no admin rights, claude-science sandbox approve can approve your own copy of the tools, for example from conda. An approved socat lets the app start, but sandboxed code still has no network until an administrator installs socat in a system folder.

Can you run Claude Science on a rented Linux server?

Yes, in one documented way. The docs describe two separate remote setups, and they are easy to confuse.

Setup 1: run all of Claude Science on a remote server

The remote server page says Claude Science runs on "a cloud VM or a lab machine". The application and your data stay on the server, and you use the web app from your own browser through an SSH (secure shell) tunnel. A rented Linux VPS is a cloud VM of this kind if it meets the requirements above. The docs do not name any VPS provider, and they give no minimum RAM or CPU, so test your own server before you depend on it.

The server needs more disk than a local install: about 10 GB free. The first launch can use about 9 GB for its starter Python and R environments. After you install the dependencies and the app, open the tunnel from your own computer:

ssh -L 8000:localhost:8000 -L 8001:localhost:8001 you@server.example.com

Leave that terminal open. Then start the app on the server:

claude-science serve --no-browser

It prints a sign-in link that starts with http://localhost:8000/?nonce=. Open the link in your own browser. The link is single-use and expires after about three minutes, so set up the tunnel first. claude-science url prints a fresh link at any time. If the sign-in redirect cannot return through the tunnel, choose Paste code instead.

Port 8000 is the web app, and port 8001 serves previews of generated HTML. If you forward only 8000, interactive previews show as static snapshots, because the preview port is not reachable. By default the web app listens only on the server's localhost, so it is not exposed to the network. To run it in the background, use claude-science serve --no-browser --detached. claude-science status tells you whether it is running, and claude-science stop stops it.

Setup 2: keep the app local and send jobs over SSH

In this setup, Claude Science stays on your computer and runs jobs on a machine you reach over SSH. The remote compute page names a lab workstation or an HPC (high-performance computing) login node, for example a workstation with a GPU or your existing cluster. Claude Science uses your existing ~/.ssh/config and authenticates with your key or ssh-agent. It installs nothing on the host.

• Claude submits Slurm jobs by default when sbatch exists on the host. Slurm is a job scheduler for clusters. Without sbatch, jobs run as detached processes.
• PBS and LSF, two other schedulers, are not supported.
• The default job timeout is 30 minutes, and the default limit is 100 concurrent jobs.
• Files over about 100 MB stay on the host.

Remote jobs run outside the sandbox, as your user. They can reach everything your account can read and write. For this reason, use an account with only the access the job needs.

That page does not mention cloud VMs or a VPS. The remote server page does list "Use the server as an SSH host" as an option for shared servers where you have no admin access, but that is about a generic server. The docs do not confirm a rented VPS as an SSH compute host. If you rent a server and have root, Setup 1 is the path the docs cover by name.

How do Team and Enterprise admins turn on Claude Science?

Pro, Max and Ultra need no admin action. Claude Science is off by default for Team and Enterprise organizations. An Owner or Primary Owner turns it on at Organization settings > Claude Science, with the Enable for your organization toggle. The Admin role can add members and assign seats, but it cannot turn Claude Science on. Until the organization enables it, members are stopped at sign-in with a Request access option.

On Enterprise, custom roles need the Claude Science capability, and the "All generally available" option does not include it. The admin controls page also sets different defaults for each plan:

• SSH hosts: on for Team, on for Enterprise.
• Modal: on for Team, off for Enterprise.
• Auto mode: on for Team, off for Enterprise.
• Custom connectors: on for Team, off for Enterprise.

If an admin switches SSH hosts off, existing hosts stay listed but refuse new commands and file transfers. Organizations with HIPAA (Health Insurance Portability and Accountability Act) compliance enabled have app access off by default. Use of the app is not covered by the BAA (business associate agreement), so keep protected health information out of it. For the plan itself, see Claude Team plans for small businesses.

What is the Claude Team plan for scientists?

It is a discounted Team plan for academic and nonprofit research groups, with Claude Science on by default. The program page showed these terms on 5 November 2026:

• Standard seats: $0 per month (regularly $20).
• Premium seats: $15 per month (regularly $100).
• Group size: 1 to 25 seats per group. Larger groups must contact Anthropic.
• Promotional period: 12 months from sign-up. After that, seats renew at the then-current price.
• Eligibility: principal investigators (PIs) or equivalent at accredited universities and nonprofit research institutes, in the natural sciences, mathematics, computer science, engineering and related fields.

For-profit companies, contract research organizations and industry R&D (research and development) teams are sent to the normal Team and Enterprise plans. The page says most applications are reviewed in 5 to 7 business days. Groups that are inactive for 90 days may move to standard Team pricing after an email. The plan includes everything in Claude Team, including Claude Code and Cowork.

Can you use Claude Science for clinical or diagnostic work?

No. Anthropic's statement is direct: "Claude Science is a research tool and isn't intended for clinical or diagnostic use."

The same warning says Claude can make mistakes. The reviewer reduces errors but does not remove them all, because it checks claims against the execution record and does not re-run the analyses. Anthropic tells you to verify results before you rely on them in research, publication or later decisions. Treat the provenance record as a way to check the work, and do that check yourself.

FAQ

Is Claude Science free?

No. Claude Science is not available on the Free plan. It is included with nothing separate to buy on the Pro, Max, Ultra, Team and Enterprise plans. The one $0 route is a Standard seat on the Claude Team plan for scientists, which is for principal investigators or equivalent at accredited universities and nonprofit research institutes. That price is promotional and lasts 12 months from sign-up.

Does Claude Science use the same limits as Claude Code?

Yes. Anthropic's docs say Claude Science usage counts toward the same usage limits as the rest of your Claude plan, including Claude Code and Cowork. A long analysis therefore reduces what is left for your coding work until the limit resets. Settings > Usage in Claude Science shows how much you have used and when each limit resets.

Does Claude Science work on Windows?

Yes, on Windows 11 x64. Anthropic's launch post from 30 June 2026 listed only macOS and Linux, but the changelog shows Windows support arrived in version 0.1.47 on 10 September 2026. The other supported systems are macOS 13 or later and Linux x64 on a glibc-based distribution.

Can I run Claude Science on a VPS?

The docs support running all of Claude Science on a remote Linux server, which they describe as a cloud VM or a lab machine. You reach it from your own browser through an SSH tunnel. The server needs x64 glibc Linux, about 10 GB of free disk, bubblewrap 0.8.0 or later, socat and unprivileged user namespaces. The docs name no VPS provider and give no minimum RAM or CPU, so test your own server first.

What happens if I reach my weekly limit during a long analysis?

Without usage credits or extra usage, the session pauses. If the reset is more than 8 hours away, it pauses for up to 8 hours, then stops and shows the reset time. After the reset, select Resume or send a message. With usage credits or extra usage on, Claude Science asks by default whether to keep going and waits for your answer. Your conversation and artifacts are kept in each case.