Lesson 1: What Claude Science Is and Is Not

Before you start: what this course is

This course is an independent guide. It is not written, reviewed or endorsed by Anthropic. Claude Science is in beta, so names, plans and limits can change. Treat Anthropic's current pages as the authority:

• The documentation at claude.com/docs/claude-science
• The product and download page at claude.com/product/claude-science

If this course and those pages disagree, the pages win.

What Claude Science is

Claude Science is a desktop application from Anthropic, launched in beta on 30 June 2026. It pairs the Claude models with an analysis environment on your own computer. It is a separate install from the ordinary Claude desktop app, and it is not a new model.

The working loop has four parts:

1. You describe a research task in plain language.
2. Claude writes Python, R or shell code and runs it in a sandbox, a restricted area on your machine.
3. Results (figures, tables, reports) are saved as versioned artifacts, each with a provenance record of what produced it.
4. A background reviewer can check Claude's claims against the work that actually ran.

A useful comparison is a capable research assistant who keeps a complete lab notebook. The assistant does the coding and records every step. You remain the investigator who signs off on the result.

The app is not limited to biology. Many of the built-in database connectors serve the life sciences, but the core loop works in any quantitative field.

Example. An energy economist asks for a regression of hourly electricity demand on temperature. Claude writes the code, runs it, and saves a coefficient table and a residual plot as artifacts. Each artifact can be traced back to the code that made it.

What it is not

Keep three limits in mind from day one.

• It is not a clinical or diagnostic tool. Anthropic describes it as a research tool that is not intended for clinical or diagnostic use. Do not use its output to diagnose or treat a patient.
• It is not a guarantee of correctness. Claude can make mistakes. The reviewer reduces errors but does not remove them. It compares claims with the record of what ran, and it does not re-run the analysis.
• It is not a finished product. Beta means features, names and limits may change.

What you should do: verify any result before it enters a paper, a report or a decision. In the regression example, that means the following checks:

• Confirm the row count matches your source file.
• Confirm the units of each variable.
• Decide for yourself whether a linear model is the right method.

The third check is yours alone. A record showing that the code ran does not show that the method was the right choice.

Check your eligibility

Work through these steps before you download anything.

1. Confirm your plan. You need a Claude account on a Pro, Max, Team or Enterprise plan. The Free plan does not include Claude Science. There is nothing separate to buy on the paid plans.
2. Confirm your operating system. The current requirements are:
   • macOS 13 or later, on Apple silicon or Intel
   • Windows 11 (x64)
   • Linux x64 on a glibc-based distribution, which has extra sandbox prerequisites covered in the next lesson
3. Confirm free disk space. Allow about 5 GB for the runtime and starter environments.
4. Confirm your region. The app is offered in the countries and regions where Anthropic offers claude.ai.
5. On Team or Enterprise, confirm it is switched on. Claude Science is off by default for these organisations. An Owner or Primary Owner must enable it in Organization settings. The Admin role cannot do this. On Pro and Max, no admin action is needed.

Two notes on these checks:

• Windows. The launch announcement and at least one help article list only macOS and Linux. The current documentation and the download page also list Windows. Check the download page for your platform.
• Blocked sign-in. If your organisation has not enabled the app, sign-in stops with a message telling you to ask your admins. Send the request to an Owner, because an Admin cannot enable it.

If you work at an academic or nonprofit research institution, Anthropic also offers a discounted Claude Team plan for scientists. Check its current terms on Anthropic's site.

How usage is counted

Claude Science does not have its own allowance. Its usage counts toward the same limits as the rest of your Claude plan, including Claude Code and Cowork. A long analysis in the morning leaves less for other Claude work that afternoon.

On plans with usage limits, two limits apply:

• A limit that resets every 5 hours
• A weekly limit

As you approach either one, the app asks whether the session should continue. If a session pauses or stops at a limit, your conversation and artifacts are kept.

You pay more than your plan price only in specific cases:

• You turn on usage credits (Pro and Max), or your admin turns on extra usage (Team).
• Your organisation is on a usage-based Enterprise plan.
• You connect an outside service that bills you directly on your own account.

What you should do: plan heavy analyses with the shared limit in mind, and learn where the meter is.

Your step for this lesson

You have not installed the app yet, so this step is a written pre-flight record.

1. Open your Claude account settings and write down your plan name.
2. Write down your operating system and version.
3. Check that you have about 5 GB of free disk space.
4. On Team or Enterprise, ask an Owner whether Claude Science is enabled, and record the answer.
5. Open Anthropic's current requirements page and compare it with your notes. Record any difference from this lesson.

Once the app is installed in the next lesson, complete the check of the record:

1. Open Settings.
2. Select Usage.
3. Read how much you have used and when each limit resets.

Key takeaways

• Claude Science is a beta desktop app from Anthropic for macOS, Windows and Linux, installed separately from the Claude desktop app.
• Claude writes and runs Python, R or shell code in a sandbox and saves results as versioned artifacts with provenance.
• It is a research tool and is not intended for clinical or diagnostic use.
• The reviewer reduces errors but does not re-run analyses, so you must verify results yourself.
• You need a Pro, Max, Team or Enterprise plan. Free is excluded, and Team and Enterprise need an Owner to enable the app.
• Usage counts against your plan's shared limits, and Settings > Usage shows your use and reset times.
• This course is independent of Anthropic, so check every detail against Anthropic's current pages.