Anthropic already had Claude writing code. Now he has put it in the laboratories

Anthropic had already placed Claude in one of the most everyday and valuable tasks in the technology industry: writing code. Now he wants to take it to more delicate terrain and with potentially much greater consequences: scientific work within laboratories. The company has introduced Claude Sciencea product designed to help researchers move between literature, data, specialized tools and computing resources.

Claude to science. The key to Claude Science is not only that Anthropic has added more tools to Claudebut in the type of problem it is trying to solve. In science, a huge part of the work involves jumping between databases, files, code, figures, citations, and computing resources that rarely talk to each other comfortably. The company wants to integrate all this into a specific application, available from June 30, 2026 in beta for Pro users, MaxTeam and Enterprise on macOS and Linux.

A category jump. Anthropic had already begun to bring Claude closer to scientific work last fall, when it launched connectors and functions under the umbrella of Claude for Life Sciences. This helped the model to relate better to software and scientific databases, but it still had a more limited scope. What is happening now goes one step further. Anthropic seems to want science to stop being just a use case and become a product line.

Verifiable work. The promise of Claude Science is not limited to helping you write or summarize. Anthropic claims it can analyze scientific literature, execute multi-step investigations, generate figures and manuscripts, and allow the researcher to refine them iteratively. The most important part is how it leaves a trace: each result includes the code, the environment, and the message history that produced it. In addition, a review agent checks quotes and calculations, and can point out untraceable numbers or figures that do not match the code that generated them.

Claude Science’s ambition might sound very broad, but his first steps have a fairly recognizable accent. Anthropic has prepared it with more than 60 capabilities and connectors targeting areas such as genomics, proteomics, structural biology, computational chemistry, and single-cell analysis.

The computation, within the flow. Many investigations do not stop at reading articles or generating figures: they also require carrying out heavy work on machines prepared for it. Anthropic says Claude Science can help prepare those processes on the researcher’s laptop, on a Linux machine, on an HPC access node via SSH, or with on-demand computing in Modal. The company clarifies that the system writes a plan and asks permission before accessing new resources, so that the researcher can review or revoke decisions. It also states that large or sensitive data can remain in the lab infrastructure, sending Claude only the context necessary for each step of the analysis.

Anthropic accompanies the launch with examples. Manifold Bio, dedicated to the design of drugs aimed at specific tissues, used Claude Science to propose targets in its experiments, evaluating surface expression, cell trafficking and safety according to the company’s own criteria. The Allen Institute used it to build a computational review template with about 20 custom skills, capable of reading thousands of articles and organizing findings into an evidence base. And at UCSF, epidemiologist Stephen Francis says the tool sped up glioma analysis to about one-tenth the time before, with results independently validated by his group.

Images | Anthropic

In Xataka | South Korea has a plan to dominate in memory chips and robotics. One of a billion dollars

Leave your vote

Leave a Comment

GIPHY App Key not set. Please check settings

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.