Anthropic Releases Claude Science, a Digital Workshop for Researchers

Anthropic has launched a public beta of Claude Science, a desktop app for macOS and Linux designed to replace the scattered mix of notebooks, scripts and databases that biologists, chemists and geneticists rely on every day. The tool isn't a new language model, but an application layer built on top of the already available Claude Opus 4.8, tailored specifically for lab and analytical work.

Claude Science's architecture splits the work between two agents. The first, a coordinator, takes a plain-language instruction and breaks it down into concrete analytical steps, choosing the right tools and databases on its own. The second, a reviewer, checks the results step by step, verifying citations, figures and generated charts before they make it into the final report.

How it works in practice

The app can run locally on a researcher's own machine or remotely, connecting over SSH to the login nodes of the computing clusters used by universities and research institutes. That matters for labs that already have their own HPC infrastructure and don't want to move sensitive research data to an external cloud.

Every chart or figure the app generates comes with a full record that lets another scientist reproduce the result: the exact code, a description of the computing environment, a plain-language explanation, and the complete exchange of messages with the agent. Anthropic says this addresses one of science's chronic problems, the difficulty of replicating someone else's results.

Early testers

Early testers included a neuroscience team at the Allen Institute and the UCSF Brain Tumor Center, which independently verified Claude Science's results in germline variant analysis, confirming that the app produces fast and reliable analyses. Biotech startup Manifold Bio said the tool's edge over a plain coding assistant is that it works end to end, from gathering the right data to applying sound judgment informed by the context of previously built analyses.

What set Claude Science apart from a plain coding assistant was its ability to work end to end, gathering the right data and applying sound judgment informed by the context of previously built pipelines - Manifold Bio

A race for the lab market

Anthropic's move is part of a broader race among large model providers for the scientific tools market. Google DeepMind is developing its own biology models, and Nvidia has spent months investing in BioNeMo as a platform for pharmaceutical companies. Claude Science differs from these offerings in that it doesn't sell access to a single model, but an entire application with a ready-made workflow, lowering the barrier to entry for smaller research teams without their own data engineers.

For Polish universities and institutes that have long relied on national supercomputers at Cyfronet or the Poznan Supercomputing and Networking Center, the ability to connect Claude Science over SSH to their own HPC nodes means the tool can be deployed without moving patient data or sequencing results outside university infrastructure. That matters given Poland's especially strict rules on protecting medical and genetic data.

Anthropic's grant program, though modest by global standards (50 projects, $30,000 in credits each), could be a real opportunity for smaller research teams in Central Europe that rarely make it onto the radar of major American sponsorship programs. The application deadline is July 15, with results expected by the end of the month.

Anthropic has not yet disclosed figures on the number of active Claude Science users or when the app will leave beta. The company says only that it plans to keep expanding the list of supported databases and scientific domains, suggesting genomics and chemistry are just the first stage.

Sources: Anthropic Newsroom (anthropic.com), MarkTechPost (marktechpost.com), TechTimes (techtimes.com)