Anthropic Launches Claude Science, Its New Flagship AI Tool for Scientific Research

Anthropic on Tuesday formally launched Claude Science, a major new product unveiled at an event attended by pharmaceutical executives, biotech startup founders, and academic researchers. The product is designed with scientific research as its core mission — mirroring the role Claude Code plays in software engineering.

What Claude Science Does

Like Claude Code, Claude Science is built for a high degree of autonomy, capable of completing meaningful research tasks from brief, high-level instructions. It integrates a suite of research-specific tools with particular utility in computational biology and drug development.

The product is now available to all paid Claude subscribers. Anthropic also announced it will deploy Claude Science internally to pursue drug candidates for rare and neglected diseases.

This is not Anthropic's first move into AI-powered science. In October last year, the company released a set of plug-in tools under the name "Claude for Life Sciences" to help Claude interact with scientific software and databases. Claude Science, however, is a fully standalone product — not an experimental add-on. By elevating it to stand alongside Claude Code and Claude Cowork, Anthropic is signaling — at minimum publicly — a serious commitment to AI in scientific research.

"This represents how important this is to our mission," said Eric Kauderer-Abrams, Anthropic's head of life sciences. "It sits alongside Claude Code and Claude Cowork as the next truly major product we're launching. Our mission is to develop AI for the long-term benefit of humanity, and we believe the life sciences represent the single greatest opportunity to do that."

Taking Aim at Google DeepMind's Scientific Lead

For much of the past decade, Google DeepMind has been the standard-bearer for AI in science. CEO Demis Hassabis and researcher John Jumper shared the Nobel Prize in Chemistry for their work on the AlphaFold protein-structure model, and DeepMind has posted major advances across meteorology, materials science, and beyond. Yet in recent months, the rapid pace of frontier AI development appears to have left DeepMind on the back foot — particularly in software coding, now the most commercially valuable AI application, where it has found itself playing catch-up.

Anthropic is well-positioned to claim the scientific AI mantle. CEO Dario Amodei holds a doctorate and is a scientist by training — a stark contrast to OpenAI's Sam Altman, who comes from a business background. Many scientists are already loyal users of tools like Claude Code: modern scientific research inevitably involves programming, yet not every researcher is a trained software engineer, making tools like Claude Code genuine productivity multipliers.

Perhaps most significantly, Anthropic recently secured a major endorsement from the scientific world: earlier this month, John Jumper announced he was leaving DeepMind to join Anthropic.

Research Capability on Par with a Graduate Student

Since AI agents powered by large language models — including Anthropic's Opus series — began demonstrating genuinely autonomous, useful capabilities in late 2025, scientists have taken note. Harvard physicist Matthew Schwartz wrote on Anthropic's official blog that, based on his hands-on experience with Claude Code and other Anthropic tools, the Opus 4.5 model performs scientific project work at roughly the level of a second-year graduate student.

Augmenting, Not Replacing, Existing Workflows

Kauderer-Abrams was clear that Claude Science is not intended to replace Claude Code or Claude Cowork in a scientist's workflow, but to extend what researchers can already do with familiar Anthropic products. For example, Claude Science can not only write code but also help scientists run it on high-performance computing clusters — a critical need for many researchers that is notoriously tedious to manage. The product also places a strong emphasis on reproducibility, allowing scientists to trace the origin of any figure or result and verify its accuracy.

While Claude Science is theoretically applicable across all scientific disciplines, it is clearly designed and marketed for molecular biology, cell biology, and drug discovery in particular. It integrates with tools spanning genetics, chemistry, and protein biology — making it highly practical for researchers searching for new therapeutics. At Tuesday's launch event, Alexander Tarashansky, who led Claude Science's development, gave a live demonstration of the system autonomously identifying novel drug candidates for phenylketonuria (PKU), a rare inherited metabolic disorder.

Doing Its Own Drug Research — for Both Altruism and Business

Rather than leaving all research to pharmaceutical companies and university labs, Anthropic plans to use Claude Science to conduct its own drug candidate research for neglected diseases — both to advance science and to gain direct insight into how Claude Science performs in real-world conditions.

Focusing on drug development when building a general-purpose scientific research tool carries obvious humanitarian rationale, and AI industry leaders frequently cite curing disease as one of the technology's most important potential contributions. It is also worth noting, however, that pharmaceutical companies command far greater budgets than academic research institutions.

Anthropic has said it is approaching its first profitable quarter, and landing major pharma contracts could help sustain profitability once the current surge in AI token consumption — sometimes called "tokenmaxxing" — begins to plateau. That financial resilience would carry added significance as the company advances toward a planned IPO.

原文來源： 查看原文