Anthropic has launched Claude Science, an application that pulls a researcher’s scattered tools into one place and lets AI agents run large parts of the work. It is the company’s biggest push yet into the scientific laboratory environment. On June 30, 2026, Anthropic announced that Claude Science is now available in beta. The company calls it an AI workbench for scientists, designed to pull together the databases, code tools, and compute that researchers juggle every day. An AI agent then moves seamlessly between them, automating many tedious and repetitive tasks.
The pitch targets a genuine pain point. Scientists work across dozens of databases, each with its own schema. They constantly switch between platforms such as PubMed, Jupyter, R, and a cluster terminal, and they wrangle file formats that require custom pipelines. Claude Science folds those steps into a single integrated environment. It can analyze the existing literature, run multistep analyses, and refine figures and manuscripts until they are ready for publication. This represents a fundamental shift in how research can be conducted, moving from disjointed manual workflows to a unified AI-driven process.
Importantly, Claude Science is not a new model. It operates using the same Claude models already available for purchase, including Opus 4.8, with no special access or enhanced capabilities. As industry observers have noted, the bet is on workflow innovation rather than raw model power. This approach acknowledges that many scientific breakthroughs are hindered not by the lack of powerful AI, but by the complexity of integrating various tools and data sources. By focusing on workflow, Anthropic aims to remove friction that currently slows down research.
An agent that shows its work
At the core of Claude Science sits a coordinating agent. It draws on more than 60 curated skills and connectors specifically set up for fields such as genomics, proteomics, structural biology, and cheminformatics. The agent can spin up other agents, including specialist ones built by the user, to handle specific sub-tasks. A separate reviewer agent checks citations and calculations, then flags and corrects errors as it proceeds. This multi-agent architecture ensures that complex research tasks are broken down into manageable pieces, with each part handled by the most appropriate tool or model.
Anthropic is leaning heavily on reproducibility, the issue that haunts modern science. Every figure produced by Claude Science arrives with the exact code and environment that generated it. Each figure also carries a plain-language explanation of how it was made, plus the full message history. This means a researcher can return months later and trace any result back to its origin, understanding exactly what steps were taken. They can also edit a figure using plain English commands. For example, asking the agent to drop gridlines or switch an axis to a log scale triggers a rewrite of the underlying code, making figure refinement accessible even to those without deep programming expertise.
The reviewer agent serves a second critical purpose. AI models are known to invent citations and numbers, a phenomenon often called hallucination. The system inspects outputs for untraceable figures and references that do not match the code. It is designed to catch its own mistakes before a human reviewer does, thereby increasing trust in AI-assisted research. This built-in quality control is essential for scientific credibility and acceptance.
It runs where the data already lives
Claude Science is built to sit on a lab’s own machines, addressing both data privacy concerns and computational needs. It works locally on macOS or Linux, or on a remote box over SSH or an HPC login node. Large jobs, such as folding a protein or running a genomics pipeline, are handled by the agent. It drafts a plan and asks for approval before reaching new resources. Then it submits the job to the lab’s own cluster, or to a Modal account for compute on demand. The work can scale from one GPU to hundreds, providing flexibility for projects of all sizes.
That design also answers a significant privacy worry. Because the app runs on the lab’s infrastructure, large or sensitive datasets never have to leave the premises. Only the context needed for each step is sent to Claude. Researchers can fork a session to compare two approaches without losing the original, enabling parallel experimentation and version control. This architecture is particularly important for fields like biomedical research, where patient data may be protected by regulations such as HIPAA or GDPR.
The launch includes a strategic tie-up with Nvidia. Claude Science uses the chipmaker’s BioNeMo Agent Toolkit to reach life-sciences models such as Evo 2, Boltz-2, and OpenFold3. It also draws on more than 60 scientific databases, including UniProt, PDB, and ChEMBL. Nvidia has spread its investment across the AI industry, and life sciences is one more frontier where their hardware and software tools are being applied. This partnership provides Claude Science with access to state-of-the-art domain-specific models that can handle tasks like molecular docking or sequence alignment.
What the early users say
Anthropic points to three beta users who have already found value in the tool. Manifold Bio, a company designing medicines that target specific tissues, used Claude Science to nominate targets for its latest experiments. They weighed factors such as surface expression, trafficking, and safety. The firm noted that the draw was that the app could run the task end to end, with the context of past programmes built in, saving weeks of manual work.
Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent template of about 20 custom skills to write long-form reviews. Sub-agents read thousands of papers, pulled the key findings, and stored them in a database, then drafted the review section by section. Lecoq said a single review used to take his team as long as two years. He now has about 10 of them completed, many running past 100 pages. This dramatic acceleration could transform how scientific literature is synthesized, enabling researchers to keep pace with the ever-growing volume of publications.
That number also highlights a potential downside. A tool that turns a two-year review into a batch of ten could speed real synthesis, but it could also flood an already strained literature with machine-generated papers. Anthropic’s answer is the reviewer agent and mandatory human checks. Stephen Francis, an epidemiologist at the UCSF Brain Tumor Center, said his glioma analysis ran in about a tenth of the usual time, and that his group verified the results manually and confirmed they held up. This validation step is crucial to maintain scientific integrity.
A high-stakes bet on the lab
The launch fits a wider plan for Anthropic. The company has framed Claude as a tool that can do real research, not just chat. Science is a market where that claim can be rigorously tested. It is also a commercial move. Anthropic is racing to win paying customers ahead of a planned public listing, and it has set out huge revenue targets to justify its spending. The company has reported strong growth in enterprise adoption, and Claude Science is positioned to capture market share in academic and pharmaceutical sectors.
The timing is awkward in one respect. Anthropic is in a tense standoff with Washington, after the US government moved to block foreign access to its most powerful models. A product built for open scientific collaboration lands in the middle of that geopolitical fight. The company must navigate export controls and ensure compliance while still encouraging global scientific exchange. This tension may affect how widely Claude Science is adopted outside the United States.
Claude Science is in beta on macOS and Linux for Pro, Max, Team, and Enterprise plans, with discounted seats for academic and nonprofit labs. Anthropic will also fund up to 50 research projects with up to $30,000 in credits each. Applications are open until July 15, 2026. The bigger question is whether AI can truly speed discovery, or simply produce more of it. The labs now testing the app will give the first real answer, and the scientific community will be watching closely to see if this new workbench lives up to its promise.
Source:TNW | Anthropic News