Anthropic's Claude Science Turns the Chatbot Into a Research Workbench

# Anthropic's Claude Science Turns the Chatbot Into a Research Workbench

Anthropic has stopped treating the laboratory as just another use case for its chatbot. With the launch of Claude Science, unveiled June 30 at a San Francisco event and now rolling out through July, the company is betting that scientists want less a conversational assistant and more a workbench—a single environment that swallows the tangle of databases, terminals, and file formats that make computational research so tedious. And in a move that signals just how far its ambitions run, Anthropic said it will begin developing drugs of its own.

Claude Science is not a new model. It runs on Anthropic's existing architecture, including Claude Opus 4.8, and layers on top a "generalist coordinating agent" with access to more than 60 curated skills and connectors pre-configured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. Those agents can spin up sub-agents and call on specialist tools, while a dedicated reviewer agent checks citations and calculations, flagging and correcting errors as a pipeline runs.

The pitch addresses a real grievance. "Scientific research is often tedious," Anthropic wrote in its announcement, describing researchers who must shuttle between PubMed, Jupyter, R, and cluster terminals, each with its own schema and bespoke data pipeline. Claude Science pulls those fragmented tools into one place where a scientist can analyze literature, run multi-step analyses, render 3D protein structures and genome browser tracks natively, and refine figures and manuscripts toward publication. Crucially, every output carries an auditable history—the exact code, the environment that produced it, and the full message trail—so results can be validated and reproduced months later.

Available now, in beta, across four tiers

The app launched in beta on macOS and Linux for Claude Pro, Max, Team, and Enterprise subscribers; Team and Enterprise admins must switch it on. It can run locally on a laptop, over SSH to a lab's own HPC cluster, or against a Modal account for on-demand GPU compute, so large or sensitive datasets never have to leave the systems they already sit on. Anthropic also introduced a discounted Team plan for academic labs and nonprofits, and is funding up to 50 "AI for Science" projects with as much as $30,000 in Claude credits each, plus up to $2,000 in Modal compute; applications ran through July 15.

Early users describe meaningful speedups. Anthropic says a neuroscientist at the Allen Institute built a multi-agent "computational review template" that produced review papers in a fraction of the two years such work once took, while a UCSF epidemiologist ran germline analyses of brain-tumor data in roughly a tenth of his previous time and independently validated the results.

The more striking announcement was strategic. Alongside the product, Anthropic said it will stand up its own preclinical drug-discovery programs aimed at "neglected" diseases—rare and tropical conditions that the pharmaceutical industry has largely written off as commercially unattractive. Eric Kauderer-Abrams, who leads Anthropic's life sciences effort, said the company wants to learn firsthand how to build the right tools and models for pharma. The push follows Anthropic's roughly $400 million acquisition of Coefficient Bio in April, which supplied the life-sciences expertise the launch clearly required.

Independent researchers reacted with cautious enthusiasm. Michael Pollastri, a Northeastern chemist who repurposes drugs for tropical diseases, told Northeastern Global News that "if Claude Science is able to automate so much of the information gathering … and help inform the ultimate decisions about where to go next, it would increase the pace of our experimentation by orders of magnitude." Jared Auclair, who works on cell and gene therapy, was more measured, warning that a general-purpose model "can hallucinate or miss nuance in regulatory guidance or assay design—errors that carry real consequences." His verdict: "It's not a shortcut to discovery—it's a co-pilot that requires a skilled pilot."

Why it matters

Claude Science lands squarely in the industry's fastest-moving new front: AI for science. OpenAI drew first with GPT-Rosalind in April, a model pitched explicitly at accelerating research and drug discovery, and Google DeepMind has spent years building scientific credibility from AlphaFold outward. Anthropic's answer reframes the contest. Rather than ship a bigger science model, it is productizing agentic tool use—orchestration, reproducibility, and compute management—and wrapping it in a workflow scientists can adopt without abandoning the tools they trust.

That is also a monetization play. By slotting Claude Science into its existing Pro, Max, Team, and Enterprise tiers, Anthropic converts a research-credibility story into recurring enterprise revenue and plants itself inside pharma and academic labs, some of the highest-value, stickiest customers in software. The internal drug program pushes the logic further: if the tools work, the payoff is not just subscriptions but a stake in the therapies themselves.

The risks are the ones Auclair named. Regulated drug development runs on validated, verifiable workflows, and a coordinating agent that hallucinates a citation or misreads an assay is a liability. Anthropic's reviewer agent and auditable-artifact design answer that objection directly—but they are promises that will be tested in real labs.

What to watch

Watch which of the 50 funded projects produce publishable, reproducible results, and whether Team and Enterprise admins actually enable the tool at scale. Watch for Windows support and broader database connectors, both early friction points flagged by users. And watch Anthropic's own neglected-disease pipeline: the first credible preclinical candidate it advances would be the clearest signal yet that the workbench is more than a well-designed interface.