Claude Science is Anthropic’s most direct move yet into specialized scientific software. The new app, now available in beta for macOS and Linux, is built as an AI workbench for researchers who need to move between literature, databases, code, figures, manuscripts, compute resources, and lab workflows without treating every step as a separate tool.
Anthropic describes Claude Science as an app rather than a new model. That distinction matters. It runs on the same Claude models available through existing plans, but changes the environment around them. Instead of a general chatbot window, researchers get a workspace built for scientific tasks, with preconfigured tools, connectors, domain-specific skills, reproducible artifacts, and access to local or remote compute.
The app is available in beta for Claude Pro, Max, Team, and Enterprise users. Team and Enterprise accounts need an administrator to enable it. Anthropic is also offering discounted Team seats for academic labs and nonprofit research organizations, along with an AI for Science program that will support up to 50 projects with credits.
For Mac users in science, the release is notable because it brings AI closer to the desktop research environment. Claude Science can run locally on macOS, connect to a Linux machine, or work through a remote system over SSH or an HPC login node. That makes it less like a consumer AI assistant and more like a lab interface for computational work.
Claude Science Is a Workflow Bet
Claude Science is not trying to win researchers only through a larger context window or a more capable biology model. Anthropic is betting that the bottleneck in scientific work is workflow.
A typical researcher may search PubMed, download data, clean files, write Python or R code, run jobs on a cluster, inspect output, generate figures, update a manuscript, check citations, and repeat the process when a reviewer asks for changes. Each step may involve different tools, file formats, databases, and computing environments.
Claude Science tries to bring those stages into one session. Anthropic says the app can help researchers analyze literature, run multi-step analyses, generate figures and manuscripts, and trace outputs back to the code and environment that produced them. Every result carries an auditable history, including message history, code, and a plain-language explanation of how the output was created.
That reproducibility focus is essential. Scientists do not only need fast answers. They need work that can be checked, repeated, cited, challenged, and revised months later. A figure that looks polished is not enough if no one can trace the data and code behind it.
That is where Claude Science differs from a normal AI chat. It is designed to produce artifacts that researchers can inspect instead of isolated answers that disappear into a conversation.
Built for Scientific Tools and Databases
Anthropic says Claude Science comes with more than 60 curated skills and connectors across genomics, single-cell analysis, proteomics, structural biology, cheminformatics, and related fields. It can query scientific sources and tools including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, GEO, journals, preprint servers, and domain-specific models.
The app also works with NVIDIA’s BioNeMo Agent Toolkit and related life-sciences models and libraries, including Evo 2, Boltz-2, and OpenFold3. That gives researchers a way to connect Claude to specialized scientific tools rather than asking a general model to invent a path through unfamiliar databases.
The user experience is agent-based. A general coordinating agent can create sub-agents, call specialist agents, and use custom agents made by a researcher or lab. Anthropic also includes a reviewer agent designed to check citations, calculations, numbers, and figures against their source code or supporting materials.
That reviewer layer is useful, but it should not be treated as a substitute for scientific validation. AI systems can check their own work, but they can also miss errors or reinforce mistakes. The best use is as a second-pass filter that flags issues before a human expert reviews them.
For labs, the stronger feature may be reuse. Claude Science can save pipelines as reusable skills, allowing future sessions to inherit preferred lab tools and workflows. That makes the app more practical for groups that already have trusted methods and do not want AI to replace them with generic suggestions.
Mac and Lab Compute Work Together
Claude Science is designed to meet researchers where their work already runs. Anthropic says the app can operate locally on a Mac or Linux machine, connect to remote infrastructure through SSH, or work with an HPC login node. It can also use Modal for on-demand compute.
That matters because scientific data can be large, sensitive, or difficult to move. A genomics pipeline, protein-folding job, or high-volume imaging workflow may not belong in a web chat upload box. Claude Science can work inside the lab’s existing infrastructure so large or sensitive datasets do not have to leave the systems where they already live. Anthropic says only the context needed for each step is sent to Claude.
This is one of the strongest arguments for a native research workbench. Scientists need AI help, but many cannot simply upload raw datasets to a consumer assistant. Institutional rules, privacy requirements, grant conditions, medical data restrictions, and compute constraints all shape what is possible.
On Mac, Claude Science could become especially useful for researchers who already use laptops as their command center while running heavier jobs on remote servers or institutional clusters. The Mac becomes the interface. The compute happens wherever the lab trusts it.
Figures, Manuscripts, and Reproducibility
Claude Science can render scientific artifacts directly, including 3D protein structures, genome browser tracks, and chemical structures. It can also generate figures and manuscripts alongside the code that created them. Researchers can ask for edits in plain language, such as changing an axis scale or adjusting the appearance of a figure, and the agent edits the underlying code.
That could reduce one of the most tedious parts of research publishing. Figure preparation often requires repeated small changes, formatting adjustments, journal-specific requirements, and back-and-forth between analysis code and presentation tools. If Claude Science can make those edits while preserving the code trail, it becomes more than a writing assistant.
Manuscript work is another major use case. Anthropic cites researchers using Claude Science to build literature review pipelines, extract findings across thousands of papers, create evidence databases, and generate review sections with citation checks. Those workflows will need careful human oversight, especially because scientific writing depends on nuance, uncertainty, and accurate representation of prior work.
Still, the direction is clear. Claude Science is trying to make AI useful not only at the idea stage, but through the repetitive, detail-heavy middle of research.
Anthropic Moves Beyond General AI Assistants
Claude Science also shows where Anthropic is taking its business. The company has already expanded Claude into coding, enterprise work, healthcare, and life sciences. A dedicated scientific workbench gives Anthropic a vertical product that can be sold to researchers, universities, biotech companies, pharmaceutical teams, and nonprofit labs.
That is different from competing only on model benchmarks. Specialized software can be harder to replace because it connects to workflows, data, compliance needs, and institutional habits. If a lab builds skills, connectors, review templates, and compute routines inside Claude Science, the product becomes part of daily research operations.
The move also places Anthropic more directly against OpenAI, Google DeepMind, Microsoft, NVIDIA, and other companies building AI tools for science. The market is moving quickly because drug discovery, protein design, computational biology, clinical research, and literature analysis all offer high-value AI use cases.
Anthropic’s approach is workflow-first. It is not presenting Claude Science as a specialized biology model with restricted access. It is presenting it as a research environment that lets the existing Claude models use the right tools, data, compute, and audit trails.
That may make the app easier for more researchers to try, especially during the beta.
A Mac App Built for Serious Research
Claude Science is important because it brings AI into the lab as software, not only as a chat window. Researchers can use it on macOS or Linux, connect it to lab infrastructure, ask it to run analyses, inspect figures, trace outputs, and reuse workflows. The app does not remove the need for scientific judgment, but it can reduce the overhead around moving from question to analysis to publication-ready output.
For Mac users in academia, biotech, and biomedical research, the release gives Claude a more serious place on the desktop. It can sit beside Jupyter, R, terminal sessions, lab databases, manuscript drafts, and cluster jobs instead of floating outside the workflow.
The risks are real. AI-generated science needs validation, citation checks, reproducibility, data controls, and expert review. A faster research pipeline can also produce faster mistakes if labs treat the system as authority rather than assistance.
But the product direction is strong. Claude Science is not selling magic discovery. It is selling a more organized research environment. For scientists who spend as much time managing tools as asking scientific questions, that may be the more useful breakthrough.