Claude Science

What Claude Science is

Claude Science is Anthropic's AI research workbench, launched June 30, 2026 as a beta desktop app for macOS (Apple Silicon and Intel) and Linux. It sits alongside Claude, Claude Code, and Claude Design as a distinct product in the Claude family, aimed at working scientists rather than general chat users. The app pairs a generalist coordinating agent, which has access to over 60 curated scientific skills, with specialist subagents it can spin up for individual tasks, plus a background reviewer agent that flags incorrect citations, untraceable numbers, and calculation errors before they reach a manuscript. Out of the box it can query 60+ scientific databases, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO, and it ships preconfigured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.

The defining design choice is reproducibility: every figure and artifact carries the exact code, environment, and conversation history that produced it, and you can edit a figure by annotating it in plain language while the agent rewrites its own code. Analyses run in persistent Python and R kernels that survive across sessions, and larger jobs can be submitted in batch over SSH to an HPC cluster or to Modal, so big datasets stay on your own infrastructure. Anthropic is explicit that Claude Science is not a new or more capable model; it runs the same Claude models available to everyone, including Claude Opus 4.8, with no special access. The beta is included with every paid Claude plan (Pro, Max, Team, and Enterprise) at no extra charge, with no standalone price and no access on the free plan.

Where Claude Science is the strongest pick

Computational life-science workflows (genomics, single-cell analysis, proteomics, structural biology, cheminformatics) where reproducibility and citation integrity matter as much as the analysis itself, and where the lab already pays for Claude.

Pricing

No free plan. Claude Science is not available on Claude's free plan and has no standalone free version. The beta requires any paid Claude subscription (Pro $20/mo, Max from $100/mo, Team from $25/user/mo, or Enterprise), where it is included at no extra charge.

• Claude Pro: $20/mo (monthly, or $17/mo billed annually). Claude Science beta access included alongside the Claude chat app. The entry point for individual researchers; usage draws on the Pro plan's standard limits.
• Claude Max: from $100/mo (monthly subscription). Claude Science beta access with 5x or 20x Pro usage, suited to long agentic analysis sessions that would exhaust a Pro allowance.
• Claude Team Standard: $25/user/mo ($20/user/mo billed annually, minimum 5 seats). Claude Science beta for the whole lab. Anthropic offers discounted seats for active scientific labs at academic institutions and nonprofit research organizations.
• Claude Team Premium: $125/user/mo ($100/user/mo billed annually, minimum 5 seats). Higher usage per seat for labs running frequent long analyses, with the same Claude Science beta access as Team Standard.
• Claude Enterprise: Custom (annual contract, via Anthropic sales). Claude Science beta plus enterprise admin and security controls, with a dedicated sales route for life sciences organizations.

Pricing verified June 2026 from the official site. Confirm current pricing before purchase.

Best for

Research scientists, computational biologists, and bioinformaticians who already pay for Claude and want one workbench that runs analyses, queries scientific databases, fact-checks citations, and keeps every figure reproducible from raw data to manuscript.

Key features

• Generalist coordinating agent with access to over 60 curated scientific skills that delegates work to specialist subagents
• Queries 60+ scientific databases out of the box, including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO
• Background reviewer agent that flags incorrect citations, untraceable numbers, and calculation errors
• Full reproducibility: every figure and artifact carries the exact code, environment, and message history that produced it
• Natural-language figure editing: annotate a plot in plain English and the agent rewrites its own code
• Persistent Python and R kernels that survive across sessions
• Batch job submission over SSH to HPC clusters or Modal, keeping large datasets on your own infrastructure
• Native rendering of proteins, molecules, sequence alignments, genomic tracks, and PDFs, plus manuscript drafting in Markdown and LaTeX

Pros

• Included in beta with every paid Claude plan at no extra charge, starting from Pro at $20/mo
• Connects to 60+ scientific databases out of the box, preconfigured for genomics, single-cell, proteomics, structural biology, and cheminformatics
• Reproducibility is built in: every artifact ships with the exact code, environment, and conversation that produced it
• Reviewer agent automatically fact-checks citations and calculations before results reach a manuscript
• Persistent Python and R kernels plus SSH batch jobs to HPC or Modal keep sensitive datasets on your own infrastructure
• Credible early adopters, with researchers at MIT, UCSF, the Whitehead Institute, and the Allen Institute cited at launch

Cons

• Beta software on macOS and Linux only: no Windows build and no web version yet
• No free tier: requires a paid Claude subscription, and long agentic analyses draw on your plan's usage limits
• Launch toolkits skew heavily to life sciences; physics, social science, and other fields get far less prebuilt support
• Runs the same Claude models as the chat app, so reasoning quality is not upgraded and outputs still need human verification despite the reviewer agent
• Large-scale jobs assume comfort with SSH, HPC clusters, or Modal, a real hurdle for wet-lab scientists without computational support

Best-fit use cases

• Running a single-cell RNA-seq or genomics analysis end to end with every step logged and reproducible
• Literature work across 60+ scientific databases with a reviewer agent fact-checking citations before submission
• Protein structure, cheminformatics, and CRISPR screen design using the preconfigured domain toolkits
• Iterating on publication figures by annotating them in plain language while the agent rewrites the underlying code
• Drafting manuscripts in Markdown or LaTeX with figures whose code, environment, and history ship alongside them

FAQ

How much does Claude Science cost?

Claude Science has no separate price. The beta is included with every paid Claude plan: Pro at $20/mo ($17/mo billed annually), Max from $100/mo, Team at $25/user/mo Standard or $125/user/mo Premium (minimum 5 seats), and Enterprise with custom pricing. It is not available on Claude's free plan.

Is Claude Science free?

No. There is no free tier, and Claude's free plan does not include it. The cheapest route in is Claude Pro at $20 per month, which includes the Claude Science beta at no extra charge. Anthropic also offers discounted Team seats for active scientific labs at academic institutions and nonprofit research organizations.

What platforms does Claude Science run on?

Claude Science is a desktop app in beta for macOS (both Apple Silicon and Intel) and Linux. There is no Windows build and no web version yet. Larger jobs can be pushed off your machine through batch submission over SSH to an HPC cluster or to Modal.

Is Claude Science a new AI model?

No. Anthropic is explicit that Claude Science is not a new or more capable model for biology: it runs the same Claude models available to everyone, including Claude Opus 4.8, with no special access. The value is the workbench itself: the coordinating agent, 60+ database connections, the reviewer agent, and reproducible artifacts.

Which scientific databases does Claude Science connect to?

Anthropic says Claude Science can query more than 60 scientific databases, naming UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO as examples. It ships preconfigured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.

How does Claude Science keep research reproducible?

Every artifact ships with its history: the exact code, computational environment, and full conversation that produced it. Figures are generated with their underlying code, you can edit them by annotating in plain language, and persistent Python and R kernels survive across sessions, so an analysis can be traced and rerun from data wrangling to publication.