Anthropic’s Claude Science is betting on workflow, not a new model, to win over scientists

Anthropic on Tuesday introduced Claude Science, an AI workflow that gives scientists a single environment to conduct computational research, saving them the hassle of navigating databases, pipelines and tools.

To be clear, Anthropic says Cloud Science “is neither a new AI paradigm nor a more capable model in biology. It runs the same Cloud Science models that are already available to everyone today (including Cloud Opus 4.8), with no special access and no portal.”

The workbench is based on Anthropic’s October 2025 launch of Claude Life Scienceswhich essentially enhanced the Claude chatbot by making it better at life science tasks. Claude Science is a place dedicated to doing this work.

The launch, announced Tuesday at a press conference on AI for Science, fits into Anthropic’s broader push to be more than just a typical provider and to own the operating layer for specific industries, in the way that Claude Code has become the operating layer for software development. Anthropic is increasingly betting its growth on vertical, workflow-level products rather than just prototype capability (which can shape how it competes and prices against competitors).

Here’s how it works: A lead AI assistant acts as a kind of project manager for the scientists. It connects to more than 60 scientific databases and comes with pre-built toolkits for specific fields, such as genomics, protein structure, and chemistry. This assistant can then create sub-helpers to help divide up the work, such as a project leader who delegates tasks to specialists, or handing off work to a dedicated “expert” assistant that the user has created for their own research. The fact-checker’s separate AI then double-checks the citations and accounts before publishing anything.

This fact-checking step is important, as more AI-assisted writing results in fabricated citations and unverifiable statistics leaking into papers. However, it is still the same basic model that is self-verifying, and not an independent source of truth.

Anthropic says Cloud Science has other ways to ensure reproducibility. For example, the workbench can create shapes such as 3D protein structures and chemistry drawers along with the code that made them. Each number includes “the exact code and environment that produced it, a clear description of how it was created, and a full messaging history,” according to the company. This process also saves scientists time by allowing them to edit figures in plain language, prompting the agent to edit its own underlying code.

Another way Claude Science can save scientists time is by running the lab’s infrastructure instead of sending data to Anthropic’s servers.

Early adopters say they’ve already started enabling this. Sean Whalen, a lead scientist in machine learning and functional genomics at the Gladstone Institutes, used Cloud Science to build a genome browser from scratch in days, according to Anthropic. Jerome Leacock, a neuroscientist at the Allen Institute, used this tool to create a multi-agent computational review pipeline.

Claude Science’s launch comes two months after OpenAI faced the same problem from a different side. In April, OpenAI has released GPT-Rosalinda specialized, fine-tuned model of biological reasoning.

The difference between the two approaches is not only whether a specialized model is necessary, but also who can access it and how quickly. Rosalind was launched as an investigational trial limited to eligible institutional customers in the United States and is undergoing review for qualifications and safety. Partners like Amgen, Allen Institute, Moderna, Thermo Fisher, and Novo Nordisk received early access.

Then there’s Google DeepMind, which plays a completely different game. DeepMind actually has basic scientific models like AlphaFold and AlphaGenome, which the other two models can only use as tools. The Gemini for Science platform also aggregates these as well as over 30 other life sciences databases into one skill set.

The end result is that three very different distribution strategies are now competing for the same scientific research market: Anthropic is expanding with broad access to subscriptions, OpenAI is narrowing and closing its doors to enterprises, and Google is relying on owned and proprietary models that no one owns. How this plays out could be an early indication of how AI vendors will compete in other niche sectors such as law, finance, and engineering in the future.

Claude Science is available in beta to anyone subscribed to Pro, Max, Team, and Enterprise. Anthropic also named Novo Nordisk and Allen Institute as customer case studies, indicating that pharma organizations are already working with multiple AI vendors.

Anthropic will also support up to 50 Cloud Science projects, providing up to $30,000 in credits: “We are looking for postdoctoral and graduate projects that span domains and explore the frontiers of science, with an early focus on areas across biomedical research. Applications are open until July 15, 2026, with award notifications sent by July 31. Projects will run from September 1 to December 1, 2026.”