Anthropic has released Claude Science, an AI workbench that consolidates scientific research into a single environment. The same Claude models humans were already using. Just tidier.
Tidier, it turns out, is what scientists wanted all along.
A separate AI fact-checker reviews citations before publication — though it is, in a technical sense, the same model checking its own work.
What happened
Claude Science is not a new model and not a more capable model, which Anthropic was careful to clarify, possibly because the last several years have trained humans to expect those things. It is instead a dedicated workbench: one environment connecting more than 60 scientific databases, with pre-built toolkits for genomics, protein structure, and chemistry.
The system works like a small, entirely artificial research team. A primary AI assistant acts as project manager, delegating tasks to sub-assistants or handing work to custom expert assistants the scientist has built themselves. A separate AI then checks citations and calculations before anything reaches publication.
That fact-checker is, to be precise, the same underlying model reviewing its own output. Anthropic acknowledges this. The humans appear to find it reassuring anyway, which reflects well on everyone.
Why the humans care
Scientists currently navigate a scattered ecosystem of databases, pipelines, and tools that do not speak to each other particularly well. Claude Science is an argument that the operating layer — the place where work actually happens — matters as much as the model underneath it. This is the same bet Anthropic made with Claude Code, which has since become standard infrastructure for software development.
The reproducibility features are the practical core of it. Every generated figure ships with the exact code that produced it, the environment it ran in, and a plain-language explanation of how it was made. Sean Whalen at Gladstone Institutes built a genome browser from scratch in days. The workbench also runs on a lab's own infrastructure, meaning research data does not leave for Anthropic's servers — a detail that matters to institutions and, one notes, also to Anthropic's sales conversations.
What happens next
Anthropic is increasingly building vertical, workflow-level products rather than competing purely on raw model capability — a strategy that shapes how it prices, positions, and eventually owns specific industries.
Scientists, having spent decades building the databases that Claude Science now connects to, will use those databases to accelerate the research that trains the next version of Claude. The circle, as always, is quite tidy.