The frustration is familiar in any lab: a researcher has a stack of papers, a pile of experimental data and a question that spans both. General-purpose chatbots can summarize the papers or chart the data, but they stumble when the two have to meet. Anthropic says it has built a tool for that gap, launching Claude Science, an AI workbench designed for scientific researchers.
The product combines the pieces researchers use every day. Claude Science handles literature analysis, letting users trace citations across thousands of papers and pull out the methods, results and contradictions relevant to their own work. It supports experiment design, suggesting controls, sample sizes and statistical approaches. And it does data visualization, turning raw results into figures without a separate coding session.
Anthropic describes the launch as its first deliberate step from general-purpose chatbot toward vertical AI, a move the company has been preparing for months. The science market is the opening wedge, chosen, according to people familiar with the strategy, because it is narrow but deep: scientists have clear workflows, clear standards for what good work looks like, and a demonstrated willingness to pay for tools that save time.
The demand is well documented. Researchers have been among the heaviest users of AI tools since the arrival of large language models, using them to draft code, parse papers and summarize the literature. But the general models have known weaknesses in this setting: they lose track of long citation chains, mishandle experimental datasets with missing values or nonstandard formats, and struggle to distinguish a claim from its evidence. Claude Science is built around those specific failures.
The product leans on capabilities Anthropic has been developing across its model line. Long-context handling lets the workbench read dozens of papers at once and keep their details straight. Tool use lets it write and run analysis code, generate charts and search databases, turning the assistant from a conversation partner into a lab instrument. The interface organizes the work the way a researcher would: projects, papers, datasets and figures, rather than a single chat thread.
The competition is already present. Google’s DeepMind has pushed AI toward scientific discovery with tools for hypothesis generation, and a generation of startups, from literature tools like Elicit and Consensus to platforms that automate data analysis, have built businesses on the same demand. What Anthropic brings is a frontier model underneath the workbench, and a brand that enterprise customers in regulated industries already trust.
For Anthropic, the vertical push is also a business necessity. The company has raised enormous sums, including a $65 billion round in May that drew Samsung as a strategic investor, and it is preparing for an initial public offering. Frontier model development is expensive, and investors want to see the models converting into durable revenue. Vertical products like Claude Science carry higher prices than consumer subscriptions and create switching costs: a lab that builds its workflow around the tool is hard to dislodge.
The way the tool is meant to be used is concrete. A materials scientist studying a new alloy can feed in the relevant papers, have the workbench extract the synthesis conditions and property measurements, design a follow-up experiment with proper controls, and generate the comparison charts, all within one session. A genomics group can trace a claim about a gene’s function through dozens of studies, seeing which papers support it and which contradict it, without losing the thread of any single result.
The product is available through Anthropic’s platform, with access for individual researchers and institutions, and the company has said it will keep the workbench connected to its broader model line as new versions arrive. The pricing is set to make adoption easy in academic settings, where budgets are small, while the enterprise version carries the higher price tag that the business model needs.
The university angle matters strategically. Researchers train the next generation of engineers and scientists, and the tools they use in the lab tend to follow them into industry. Anthropic executives have described academia as a distribution channel as much as a market, and Claude Science is priced and positioned to get into labs early.
Skeptics note that scientists are a demanding audience, quick to abandon tools that produce confident errors, and that AI systems still hallucinate, especially on specialized material. Anthropic says Claude Science is designed to flag uncertainty and cite its sources, addressing the trust problem head-on. Whether that is enough will be tested in the journals, where a wrong figure costs more than a subscription.
The launch is the first of what Anthropic says will be a series of vertical products, and the science workbench is the proof of concept. If researchers adopt it, the model is simple: find a deep, technical market, build a tool around the model, and let the workflow do the selling. The scientists were already using AI; Anthropic is betting they will pay for one that understands their work.
