Anthropic has launched Claude Science, a dedicated workbench for computational research that runs on the company’s existing Claude models rather than introducing new AI capabilities. The move signals a strategic bet that scientists need better workflow tools, not more powerful models, to accelerate their work.
What Claude Science is, and isn’t
Claude Science is not a new AI model. It runs the same Claude models already available to subscribers, including Claude Opus 4.8, with no special access, gating, or fine-tuning for biology. Instead, it is a unified environment that connects a central AI assistant to more than 60 scientific databases, with prebuilt toolkits for genomics, protein structure analysis, chemistry, and other research domains.
The assistant acts as a project manager, spawning sub-assistants to handle specialized tasks in parallel. Researchers can also plug in their own custom-built expert assistants. A built-in fact-checker double-checks citations and calculations, though it runs on the same underlying model, a limitation Anthropic acknowledges.
Reproducibility built in
One of the more notable features is how Claude Science handles figures. Every generated image, a 3D protein structure, a chemistry drawing, a data plot, is saved alongside the exact code and environment that produced it, plus a plain-language description of its creation and the full message history. Researchers can then edit any figure in natural language, prompting the agent to modify its own underlying code.
The workbench can run on a lab’s own infrastructure, meaning sensitive data never leaves local servers.
Early results
Anthropic pointed to two early adopters. At the Allen Institute, neuroscientist Jerome Lecoq used Claude Science to build a multi-agent computational review pipeline. At the UCSF Brain Tumor Center, Stephen Francis’s group accelerated a comprehensive germline analysis of glioma from what previously required extensive hands-on time to a fraction of that, with results independently validated.
Funding for early-career researchers
To drive adoption, Anthropic is offering up to US$30,000 (approximately £24,000) in compute credits for postdoctoral and graduate projects, prioritizing biomedical research. Applications are open through July 15, 2026, with projects running from September to December.
Three competing strategies
Claude Science enters a scientific AI market where three very different approaches are competing for the same users. OpenAI’s GPT-Rosalind, launched in April 2026, is a specialized model fine-tuned for biological reasoning, gated behind safety reviews and limited to US enterprise customers including Amgen, Moderna, and Novo Nordisk. Google DeepMind’s various science platforms are built on proprietary models like AlphaFold and AlphaGenome, which no other company has access to.
Anthropic’s bet is that openness and workflow integration will win more scientists than a proprietary model or a narrow biological specialist. The strategy mirrors the approach Anthropic took with Claude Code for software development: build a workbench around the existing model, integrate deeply with the user’s tools, and make it broadly available.
The outcome in scientific research could serve as an early signal for how enterprise AI competition plays out in other verticals, law, finance, and engineering are watching similar debates play out.
Sources: Anthropic’s Claude Science bets on workflow (TechCrunch, June 30, 2026)