◈AI Codex
Features & Updatesupdate

Claude Science: Anthropic builds a Claude Code for the lab

In brief

On June 30, 2026 Anthropic launched Claude Science, a research workbench that pulls scientific tools — databases, compute, visualization, a citation-checking reviewer — into one place the way Claude Code did for software. It's built for genomics, proteomics, and cheminformatics labs, not general knowledge work. Here's what it is, who it's actually for, and the product pattern it signals.

6 min read·Claude

Contents

♡Sign in to save

Claude Science is a research application Anthropic launched in beta on June 30, 2026. The one-line version: it's what Claude Code is for software engineering, pointed at scientific research instead. It runs on macOS and Linux and is available to Claude Pro, Max, Team, and Enterprise subscribers.

This is a specialist product. If you're a marketer, an operator, or a developer building a chatbot, Claude Science is not for your day job — the general Claude apps and API still are. It matters to you as a signal about where Anthropic is heading, which is the last section of this piece. But first, what it actually does.

What it is

Science work is fragmented across tools: one program to pull sequences from a database, another to run an analysis, a third to make a figure, a manuscript editor to write it up, and a compute cluster somewhere to run the heavy jobs. Claude Science is a single environment where an AI agent drives all of those, with the researcher supervising.

It covers the full arc of a study:

  • Literature analysis — reading and synthesizing prior work.
  • Multi-step experiments — running analyses that chain many tools together.
  • Data visualization — with native, in-app rendering of 3D protein structures, genome tracks, and chemical structures, so you see the biology, not a generic chart.
  • Manuscript preparation — drafting the write-up.

The target fields are specific: genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.

What's under the hood

Three things make it more than "Claude with a science prompt":

60+ pre-configured scientific skills and connectors, wired to more than 60 scientific databases — UniProt, the Protein Data Bank (PDB), Ensembl, and the rest of the standard toolkit — so the agent can pull real data instead of guessing at it. Labs can add their own skills and connect proprietary internal tools.

Compute management across three tiers — a laptop for light work, an HPC cluster for heavy jobs, or on-demand GPUs through Modal — so a long-running job goes to the right hardware without the researcher hand-managing infrastructure.

A reviewer agent that checks the work. Claude Science uses actor–critic agent pairs: one agent does the analysis, a second checks citations and recalculates numbers before results are trusted. Every output is reproducible, carrying its complete code history — the auditable trail that a scientific result needs and a chatbot answer doesn't. This is the feature that separates a research tool from a writing assistant: the output is meant to survive peer review, so it ships with its provenance.

Availability and the fine print

  • Beta on macOS and Linux, for Pro, Max, Team, and Enterprise plans.
  • A discounted Team plan for academic and nonprofit research labs.
  • An AI for Science credits program offered up to $30,000 in credits for selected projects; applications for that program closed July 15, 2026.

Anthropic also said it will use Claude Science itself, to pursue research into drugs for rare and neglected diseases — the kind of work that's scientifically tractable but commercially neglected.

The pattern this signals

Claude Science is the clearest example yet of a strategy worth naming, because it will keep repeating. Anthropic is moving from one general model you prompt to vertical products that wrap the model in the tools, data, and guardrails of a specific kind of work:

  • Claude Code wrapped the model in a terminal, a filesystem, and a git workflow — for software.
  • Claude Science wraps it in scientific databases, compute, and a citation-checking reviewer — for research.

The shared shape is: take the domains where an expert spends most of their time gluing tools together, and build the glue. Expect more of these — for other professions where the work is tool-heavy and the output has to be auditable. If your organization has a function that looks like that (finance, legal, clinical), the useful question isn't "should we use Claude Science" — it's "what would the vertical Claude product for our work include, and can we assemble a version of it now with Projects, connectors, and skills?"

  • The six phases of AI adoption — where "vertical product" sits on the curve from individual use to autonomous execution
  • Claude for Creative Work — the same vertical-product move, aimed at design and creative tools
  • Ask Your Org — assembling a domain-specific Claude from connectors, without waiting for a packaged product

Source: Claude Science, an AI workbench for scientists, June 30, 2026.

Weekly brief

For people actually using Claude at work.

Each week: one thing Claude can do in your work that most people haven't figured out yet — plus the failure modes to avoid. No tutorials. No hype.

No spam. Unsubscribe anytime.

What to read next

All articles →
Claude·Field Note·7 min

Using Claude for customer support: what actually works

Customer support is the most common first AI use case for a reason — and the place where the most teams get burned. Here's what a working implementation looks like, and what the common shortcuts miss.

Claude·Role-Specific·5 min

Five habits that separate operators who get results from those who don't

The gap between people who get consistent value from Claude and people who don't isn't intelligence or skill — it's four habits most people never develop. Here's what effective Claude users actually do differently.

update·6 min

Claude Tag: a virtual coworker that lives in your Slack

Anthropic launched Claude Tag on June 23, 2026 — a research preview that puts Claude inside Slack as a shared team member. You tag @Claude into a channel, hand it a task, and it works through the steps on its own, posting progress where the whole team can see and redirect it. Here's what it does, how it differs from the Claude Slack connector you already know, and who can use it.