Science has a new assistant. Anthropic chose an event with pharmaceutical executives and researchers to unveil Claude Science, an agent based on Large Language Models that promises to support researchers in drug discovery and computational biology. This is not a simple update to the life-science plugins launched last October; it is a standalone product, placed on the same level as Claude Code and Claude Cowork. Starting today, it is available to all paid subscribers.

The autonomy of AI in the lab

At first glance, Claude Science resembles Claude Code: it receives concise instructions and works independently. But its hunting ground is the digital laboratory. The system integrates tools for genetics, chemistry, and protein biology, and can manage code execution on compute clusters—a task often daunting for non-expert software engineers. Anthropic emphasizes reproducibility: every figure or result is traceable and verifiable. During the demo, the team showed how Claude Science identified drug candidates for phenylketonuria, a rare genetic disease.

A move that reshapes the balance

This comes at a time of realignment in AI-for-science. For years, Google DeepMind led the way with AlphaFold and contributions to meteorology and materials. Then the AI frontier accelerated toward language models for coding, and DeepMind seemed to lag. Anthropic seizes the moment: CEO Dario Amodei is a PhD scientist, and the company just welcomed John Jumper, the Nobel laureate behind AlphaFold who left DeepMind. A Harvard physicist estimates that Opus 4.5, the model powering Claude Science, can execute scientific projects at roughly the level of a second-year PhD student.

Science meets data sovereignty

For those building on-premise research pipelines, the announcement raises familiar questions. Claude Science runs in Anthropic's cloud—a setting that may cause hesitation among academic labs or pharmaceutical companies holding proprietary data or facing strict compliance requirements. The AI-managed compute clusters make the product powerful but introduce trade-offs around data residency and operational costs. If the trend is toward an increasingly autonomous scientific agent, the need grows to evaluate self-hosted alternatives for those who cannot move sensitive data outside their perimeter. This is a scenario where tools like Claude Science could fuel demand for on-premise infrastructure optimized for LLMs, feeding a debate already lively on AI-RADAR.

Beyond the public good: the bottom line

Anthropic will also use Claude Science to conduct its own research on neglected diseases, but the move has a clear commercial angle. Pharmaceutical companies have far deeper pockets than academia, and new contracts could sustain the company's finances just as the “tokenmaxxing” craze fades and the IPO approaches. The convergence of LLMs and life sciences is set to become one of the hottest arenas in the coming years, and Anthropic has just made a weighty play.