Anthropic is taking a workflow-first approach to AI for research with Claude Science, a new AI workbench introduced on Tuesday. The company is positioning it as one place for scientists to run computational research without constantly moving between databases, pipelines and separate tools.
The important distinction is that Claude Science is not presented as a new model. Anthropic says it runs the same Claude models already available today, including Claude Opus 4.8, with no special access and no gating.
A dedicated workspace, not a new model
Claude Science builds on Anthropic’s October 2025 launch of Claude for Life Sciences. That earlier effort made the Claude chatbot more useful for life sciences work. The new product turns that capability into a dedicated environment for scientific tasks.
The move fits a broader strategy. Anthropic is trying to be more than a provider of general AI models. With Claude Science, it is moving further into the operating layer for a specific industry, in a similar way that Claude Code has become a working layer for software development.
That matters because the product is not framed around raw model performance alone. The bet is that scientists may value an organized research environment that can connect tools, manage tasks and keep work reproducible. In that view, the workflow around the model becomes as important as the model itself.
How Claude Science organizes research work
At the center of Claude Science is a main AI assistant that acts like a project manager for scientific work. It can connect to more than 60 scientific databases and includes prebuilt toolkits for fields such as genomics, protein structure and chemistry.
The assistant can also create sub-assistants to divide work into smaller pieces. Anthropic’s setup allows the main assistant to delegate tasks, or to hand work to a custom expert assistant that a user has built for their own research.
For scientists, the practical goal is to reduce the friction of moving between disconnected systems. Instead of using one tool for data, another for analysis, another for figures and another for documentation, Claude Science attempts to keep those steps inside one environment.
The workbench also includes a separate fact-checker AI that reviews citations and calculations before material is prepared for publication. That feature is aimed at a real problem in AI-assisted writing: fabricated citations and unverifiable statistics can slip into papers.
There is still an important limitation. The fact-checking step relies on the same underlying model checking itself, rather than an independent source of truth. That means it may help catch problems, but it does not remove the need for scientific review.
Reproducibility is part of the pitch
Anthropic is also emphasizing reproducibility. Claude Science can generate figures such as 3D protein structures and chemistry drawers together with the code that created them.
According to the company, each figure includes the exact code and environment behind it, a plain-language explanation of how it was made and the full message history. That packaging is meant to make the path from prompt to output easier to inspect.
The same setup can also make figure editing less manual. Scientists can ask for figure changes in plain language, and the agent can edit the underlying code in response. The result is a workflow where visual output and computational steps remain connected.
Another time-saving feature is infrastructure flexibility. Claude Science can run on a lab’s own infrastructure setup instead of sending data to Anthropic’s servers. For research teams, that could matter when computational work is tied closely to local systems and existing lab processes.
Early use cases and the market context
Anthropic named early users who are already applying Claude Science to research work. Allen Institute neuroscientist Jérôme Lecoq used it to build a multi-agent computational review pipeline.
Stephen Francis’s group at the UCSF Brain Tumor Center used Claude Science to speed up comprehensive germline analysis of glioma to a sliver of the time it previously required. The results were independently validated.
The launch also shows how major AI companies are taking different routes into scientific research. OpenAI approached the problem from another direction in April with GPT-Rosalind, a specialized model fine-tuned for biological reasoning.
Access is one difference between the approaches. GPT-Rosalind launched as a research preview for qualified enterprise customers in the U.S., with a qualification and safety review. Partners including Amgen, Allen Institute, Moderna, Thermo Fisher and Novo Nordisk received early access.
Google DeepMind is pursuing another strategy. It owns foundational science models such as AlphaFold and AlphaGenome, which Anthropic and OpenAI can only call into as tools. Its Gemini for Science platform combines those models with more than 30 life science databases into one skill set.
Taken together, the three approaches highlight a wider contest in AI for science:
- Anthropic is using broad subscription access and a workflow-level product.
- OpenAI is using a specialized model with enterprise-gated access.
- Google DeepMind is leaning on proprietary science models it owns.
That competition may offer an early signal for how AI vendors try to win other specialized markets such as law, finance and engineering. The question is not only which model performs best, but which company can make AI fit the way expert work actually gets done.
Who can use Claude Science now
Claude Science is available in beta to anyone with Pro, Max, Team and Enterprise subscriptions. Anthropic also named Novo Nordisk and Allen Institute as customer case studies, which suggests pharma organizations are already working with multiple AI vendors.
The company is also backing research projects through credits. Anthropic will support up to 50 Claude Science projects with up to $30,000 in credits.
The program is aimed at postdoctoral and graduate projects across domains, with an early focus on fields across biomedical research. Applications are open through July 15, 2026, with award notifications sent out by July 31. Projects will run from September 1 to December 1, 2026.
For Anthropic, Claude Science is a clear statement about where AI products may be heading. The model remains central, but the product value is increasingly in the surrounding system: the databases it connects to, the assistants it coordinates, the evidence it preserves and the workflows it makes easier to repeat.