Anthropic used Tuesday’s event for pharmaceutical executives, biotech founders, and researchers to unveil a new flagship: Claude Science. The company is framing it as an AI agent for science, not a plug-in, positioning it to support scientific research in the same way Claude Code supports software engineering. Like Claude Code, Claude Science can autonomously carry out meaningful work when given concise, high-level instructions. It also has access to tools that make it particularly useful for research in computational biology and drug development.

The practical signal here is that Anthropic is upgrading its “AI in science” approach from add-ons to a standalone product. Claude Science is now available to all paid Claude subscribers, and Anthropic says it will use the system for its own research into drugs for rare, neglected diseases. That is a meaningful shift in both capability and incentive: Anthropic is no longer only selling scientific assistance. It is also committing to run experiments with the tool itself, and then learn from the outcomes.

Claude Science is not Anthropic’s first science-related release. In October, the company released “Claude for Life Sciences,” a set of plug-ins that help Claude use scientific software and databases. The difference today is product rank and workflow expectation. With Claude Science, Anthropic is offering a full-featured standalone product, and it is putting that product alongside its other top-tier agents: Claude Code and Claude Cowork. Eric Kauderer-Abrams, Anthropic’s head of life sciences, ties that elevation directly to mission priority. He says the company views this as “right up there with Claude Code and Claude Cowork” as a “next really significant product,” and he connects the effort to Anthropic’s belief that life sciences is where it can best serve humanity’s long-term well-being.

That “where to bet” question matters for decision-makers because AI for science has been a competitive stage for a while. For roughly a decade, Google DeepMind has been widely seen as the vanguard of AI for science, with Demis Hassabis and John Jumper winning the Nobel Prize in chemistry for their AlphaFold model. DeepMind has also contributed to areas such as meteorology and materials science. But the source points out that in the past several months, fast-advancing AI progress seems to have left DeepMind “in the dust,” at least in certain usage patterns. For coding, which has become a lucrative LLM use case, the piece says DeepMind is “stuck playing catch-up.” Anthropic, by contrast, is betting that agents and frontier models can now do more than assist with text. They can execute research work.

There is also a talent and credibility storyline. The source notes that earlier this month, Jumper announced that he is leaving DeepMind for Anthropic. And it frames the broader moment: since agents powered by LLMs, including Anthropic’s Opus model series, became capable of useful, independent work in late 2025, scientists have been testing what autonomy can do in real workflows. Using work with Claude Code and other Anthropic tools, Harvard physicist Matthew Schwartz estimated in a blog post on Anthropic’s website that Opus 4.5 is about as capable of executing scientific projects as a second-year graduate student. For leaders considering vendor bets, that is the kind of claim you either discount or stress-test, but you cannot ignore it because it signals what Anthropic wants customers to expect.

Claude Science, according to the source, is designed to build on existing Claude workflows rather than replace them. It is not intended to displace Claude Code and Claude Cowork. Instead, it aims to help scientists use what they already find useful, while pushing further into “research agent” territory. The product writes code, but it also helps scientists run their code on powerful computer clusters, where many scientific teams struggle with operational complexity. It also prioritizes reproducibility, so scientists can trace the source of figures or results and check accuracy and validity. Those features are the difference between “an AI that drafts” and “an AI that can stand up in a lab meeting.”

The launch also comes with a clear focus area. Although Claude Science could theoretically assist with any scientific research, it is “designed and marketed” as a tool for molecular and cellular biology, and for drug development in particular. It can interface with tools used in genetics, chemistry, and protein biology, which aligns with how many drug discovery pipelines operate: hypothesis, computation, wet lab validation, repeat. During the event, Alexander Tarashansky, who led development of Claude Science, demonstrated how the system could autonomously identify new drug candidates for phenylketonuria, a rare genetic disease. That choice signals Anthropic’s emphasis on concrete discovery tasks rather than broad “general research” branding.

Finally, Anthropic is not limiting this to enterprise or academia partnerships. It says it will pursue its own drug-candidate research into neglected diseases using Claude Science, combining humanitarian motivation with business logic. The source highlights that pharmaceutical companies typically have far deeper pockets than academic researchers. It also notes Anthropic says it is set to see its first profitable quarter. If additional contracts with pharmaceutical companies materialize, those revenue drivers could help sustain profitability as the “tokenmaxxing craze” dies down. This is especially relevant with an IPO approaching later this year, because the market tends to reward near-term monetization proof, not just impressive demos.

For executives deciding where to allocate budgets and talent, the strategic stake is straightforward: Claude Science suggests Anthropic is treating AI-powered drug discovery like a core platform, not an experiment. If it works at scale, it could change how pharma and biotech teams structure scientific workflow automation, reproducibility requirements, and the division of labor between scientists and AI agents. And if it does not, the competitive pressure it creates still forces a reckoning across the “AI for science” vendors and internal innovation teams alike.