Anthropic has launched Claude Science, a new AI workbench designed to help researchers manage complex scientific workflows from one place. Announced on June 30, 2026, the product is now available in public beta and is aimed at scientists who need support across literature review, data analysis, coding, compute, visualization, and manuscript preparation.
The important detail is that Claude Science is not a new AI model. It runs on existing Claude models, including Claude Opus 4.8, but adds a specialized workflow layer around them. That layer connects the model to research databases, code execution, scientific tools, compute environments, and review systems.
Anthropic is positioning the product as a response to one of the biggest problems in research: fragmentation. Scientists often move between databases, notebooks, file formats, analysis tools, cluster terminals, and writing environments. Claude Science is meant to reduce that tool-switching by putting more of the research process inside a single AI-assisted workspace.
How Claude Science Works
Claude Science uses a multi-agent system. A researcher can speak to a main coordinating agent in plain language, then that agent breaks the request into smaller tasks and sends them to specialist agents. These agents can be configured for specific scientific fields or workflows, and researchers can also create their own custom agents for repeated tasks.
A separate reviewer agent runs alongside the workflow. Its job is to inspect outputs, check citations, flag numbers or claims it cannot trace, and correct issues as the work develops. That is a key part of the product because fabricated citations, weak sourcing, and unverified claims remain major risks in AI-assisted research.
The platform also includes more than 60 curated skills, databases, and connectors. These cover areas such as genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. The goal is to make Claude more useful for real scientific work, not just general question-answering.
Built for Scientific Output
Claude Science is designed to display scientific material directly inside the workspace. It can render 3D protein structures, genome browser tracks, and chemical structures without forcing researchers into separate tools. Users can also ask it to modify charts or figures in plain language, such as changing an axis scale, removing gridlines, or adjusting a visual layout.
The system can also connect to heavier compute resources. Since Claude itself cannot run large protein-folding or genomics jobs locally, Claude Science can connect to existing high-performance computing clusters or remote compute environments. It can draft a plan, request permission before accessing new resources, submit jobs, and monitor progress.
Researchers can also fork sessions to test different approaches. That makes the tool more like a research environment than a simple chatbot.
Reproducibility Is Central
One of the strongest parts of Claude Science is its focus on reproducibility. When the system generates a figure or analysis, it saves the code, environment, explanation, and message history that produced it. This allows researchers to trace how a result was created later.
That matters because scientific work depends on auditability. An AI tool that produces answers without showing how it reached them is not enough for serious research. Claude Science is trying to make AI-generated work easier to inspect, repeat, and verify.
Early Research Use Cases
Anthropic has highlighted early beta use across scientific review, biomedical analysis, and drug discovery. In one case, researchers used the system to create long structured review outputs with agent-checked citations. In another, the tool was used to speed up germline variant analysis in brain tumor research, with results independently checked afterward.
The company also described use cases where the workbench helped identify possible experimental targets from proprietary data. These examples suggest Claude Science is being positioned not only as a writing assistant, but as a system that can support real scientific decision-making.
Availability and Privacy
Claude Science is available in beta for macOS and Linux users on Pro, Max, Team, and Enterprise plans. Team and Enterprise administrators must enable access before users can work with it. Anthropic is also offering discounted access for active university and nonprofit labs.
Privacy is a major part of the pitch. The system is designed so research data can remain inside a user’s own infrastructure unless the user chooses to send it elsewhere. That is important for labs working with sensitive datasets, unpublished findings, or proprietary research.
The Bigger AI Science Race
Claude Science arrives as major AI companies compete to build tools for scientific discovery. The race is moving beyond general chatbots and into specialized workbenches that connect models to real tasks, databases, compute, and verification systems.
Anthropic’s strategy is to make existing Claude models more useful through workflow design. The message is that raw intelligence is not enough. For scientists, the bigger value comes from agents that can run processes, check their work, connect to tools, and preserve the evidence behind every result.
The launch also shows where AI products are heading more broadly. Instead of only releasing stronger models, companies are building industry-specific systems around them. For science, that means the next breakthrough may come not just from a smarter model, but from a better harness for using AI safely, transparently, and reproducibly.
Comments