Anthropic Targets Laboratories with Claude Science
Anthropic has launched Claude Science, a specialized interface built to automate complex research projects. Expanding on the company’s existing Claude Code and Claude Cowork, the platform is designed to manage computer clusters, help scientists run their code, and prioritize reproducibility, so that scientists can trace back the source of any figure or result and check it for accuracy and validity.
Automating the Scientific Workflow
Eric Kauderer-Abrams, Anthropic’s head of life sciences, says the product is intended to serve as a core component of the company’s mission to develop AI that serves humanity’s long-term well-being. Unlike general-purpose chatbots, Claude Science targets the technical demands of laboratory and computational research through three primary functions:
- Automated Coding: It writes code tailored for scientific workflows.
- Cluster Management: It assists researchers in running code on powerful computer clusters, reducing the manual overhead typically required for high-performance computing.
- Reproducibility Tracking: The system maintains a record of how figures and results were generated, allowing scientists to verify the validity of their outputs.
Recruiting Talent from Google DeepMind
Anthropic is positioning itself to compete directly with Google DeepMind, which has been at the vanguard of AI for science for the last decade. DeepMind gained significant recognition for its AlphaFold model—which earned Demis Hassabis and John Jumper a Nobel Prize in Chemistry—but recent shifts in talent suggest a changing industry.
John Jumper announced his departure from Google DeepMind to join Anthropic earlier this month. This move underscores a broader trend where research-heavy organizations are prioritizing the development of AI agents capable of independent project execution.
Grading the Model’s Scientific Competency
The efficacy of AI in science is increasingly measured by its ability to perform tasks previously reserved for human researchers. In a blog post published on Anthropic’s website, Harvard physicist Matthew Schwartz evaluated the performance of the company’s Opus 4.5 model. Based on his work with Claude Code and other Anthropic tools, Schwartz estimated that the model is about as capable of executing scientific projects as a second-year graduate student.
Comparing AI Research Philosophies
| Feature | Google DeepMind | Anthropic |
|---|---|---|
| Primary Focus | Fundamental scientific models (e.g., AlphaFold) | Agentic workflows and research productivity |
| Leadership | CEO Demis Hassabis (PhD Scientist) | CEO Dario Amodei (PhD Scientist) |
| Tooling | Large-scale predictive modeling | Integrated coding and cluster-management agents |
Bridging the Computational Gap
DeepMind has historically focused on breakthroughs in meteorology and materials science through predictive modeling. Anthropic, conversely, is targeting the daily productivity of scientists. By integrating Claude Science into existing workflows, the company aims to bridge the gap between non-expert programmers and the high-level computational power required for modern scientific discovery.