Anthropic's Claude Science: an Opus 4.5-powered AI tool for scientific research

On June 30, 2026, at a closed event for biopharmaceutical companies, biotech startups, and scientists, Anthropic unveiled Claude Science, a new flagship product that supports scientific research the way Claude Code supports software development.

What Claude Science can do

Like Claude Code, Claude Science can autonomously carry out substantive work from a short natural-language instruction, without step-by-step human supervision. The product brings together the tools, compute, and scientific databases that multi-step research workflows require in a single workspace.

• The announcement came on June 30, 2026, at an event for biopharma representatives
• Runs on the Claude Opus 4.5 model
• Connects to more than 60 scientific databases, including UniProt, PDB, Ensembl, ChEMBL, ClinVar, GEO, and others
• Available to subscribers on the Claude Pro, Max, Team, and Enterprise plans
• Anthropic will fund up to 50 "AI for Science" projects with compute grants of up to $30 000 each

The tool natively renders 3D protein structures, genomic tracks, and chemical formulas, and it can distribute compute across a laptop, HPC clusters, and on-demand GPUs. A separate reviewer agent checks citations and calculations, and results are saved as reproducible artifacts with full code history. That matters for scientific publishing, where other labs need to be able to repeat an experiment.

Claude Science runs locally on macOS and Linux, or remotely via SSH to compute clusters. Anthropic offers discounted Team plan seats for academic institutions and nonprofits.

Where Claude Science came from

The product grew out of the narrower Claude for Life Sciences plugins that Anthropic previously released for specific biology tasks. Claude Science combines them into a single standalone product with access to all the tools at once, rather than one plugin at a time. It's the same path Claude Code took when it grew from helper features into a standalone flagship product for developers.

How Claude Science resembles Claude Code

Both products are built on the same idea: give a Claude model a high-level task and let it see the work through on its own. Claude Code does this for programming; Claude Science does it for computational biology and drug discovery, including target identification, genomic data analysis, and protein structure modeling.

Harvard physicist Matthew Schwartz, one of the product's early users, assessed the capabilities of the underlying Claude Opus 4.5 model this way:

"Opus 4.5 can carry out scientific projects at roughly the level of a

second-year grad student."

Alongside the product, Anthropic announced its own internal drug development program for "neglected" diseases, which get little attention from pharmaceutical companies because of their low commercial appeal.

What it means

Anthropic is betting on the lab, not just on code. With Claude Code now a standard developer tool, the company hopes to repeat that success in science by speeding up the routine but labor-intensive stages of research, from pulling data out of scattered databases to checking whether experiments are reproducible.

FAQ

How much does Claude Science cost?

There's no separate fee: the product is included in the existing paid Claude Pro, Max, Team, and Enterprise subscriptions.

What model does Claude Science run on?

Claude Opus 4.5. The product itself is not a separate model; it's a layer on top that provides access to scientific databases and compute resources.

What do scientists whose projects Anthropic selects get?

Up to 50 projects in the "AI for Science" track will receive compute grants of up to $30 000 each, and partner Modal will add up to $2 000 more for select projects.