On June 30, 2026, Anthropic announced the beta launch of Claude Science, an AI workbench designed specifically for scientific researchers. The tool aims to solve a persistent pain point in modern research: the fragmentation of tools, databases, and compute environments that scientists must juggle daily. By integrating these disparate elements into a single platform and deploying an AI agent to orchestrate workflows, Anthropic hopes to accelerate the pace of scientific discovery while improving reproducibility.
Claude Science is not a new AI model. It runs on the same Claude models already available—including Opus 4.8—without any special access or enhanced capabilities. As noted by industry observers, the bet here is on workflow integration rather than raw model power. The company has packaged over 60 curated skills and connectors tailored to fields such as genomics, proteomics, structural biology, and cheminformatics. These pre-built components allow the system to interact with vital resources like UniProt, PDB, ChEMBL, and PubMed, as well as computational tools like Jupyter, R, and cluster terminals.
An agent that shows its work
At the heart of Claude Science is a coordinating agent that can plan and execute multi-step analyses. It can spin up specialist sub-agents, including custom-built ones created by users, to handle complex tasks such as literature reviews, data processing, and figure generation. A separate reviewer agent checks for errors—citations that don't exist, calculations that don't match outputs, and references that are fabricated—a common problem with AI-generated content. This built-in validation mechanism is designed to catch mistakes before a human reviewer has to.
Reproducibility is a central theme. Every figure produced by Claude Science arrives with the exact code, environment, and plain-language explanation of how it was created, along with the full message history. Researchers can revisit results months later and trace every step. They can also modify figures using natural language—for example, asking the agent to remove gridlines or switch an axis to a log scale—and the agent will rewrite its own code accordingly. This transparency is crucial for building trust in AI-assisted research.
Designed for local and sensitive data
Claude Science runs locally on the researcher's own infrastructure, whether that is a macOS or Linux machine, a remote server via SSH, or an HPC login node. This design addresses a major privacy concern: large or sensitive datasets never have to leave the lab. Only the context needed for each step is sent to Claude's cloud, and the rest stays on-premises. For heavy computational tasks like protein folding or genomics pipelines, the agent can submit jobs to the lab's own cluster or to a third-party compute service like Modal, scaling from one GPU to hundreds.
The platform includes the ability to fork a session, enabling researchers to compare different approaches without losing the original analysis. This feature encourages experimentation and iterative refinement, which are hallmarks of rigorous scientific work.
Partnership with Nvidia
Anthropic has integrated the tool with Nvidia's BioNeMo Agent Toolkit, giving Claude Science access to specialized life-sciences models such as Evo 2, Boltz-2, and OpenFold3. Nvidia has been investing heavily in AI for science, and this collaboration reflects a broader trend of hardware companies partnering with AI developers to target the research community. By leveraging these cutting-edge models, Claude Science can perform tasks like structure prediction, sequence analysis, and drug-target identification with state-of-the-art accuracy.
Early user experiences
Initial feedback from beta users is promising. Manifold Bio, a company designing tissue-specific medicines, used Claude Science to nominate targets for its latest experiments. The firm reported that the tool could weigh surface expression, trafficking, and safety criteria simultaneously, and that the built-in context from past projects eliminated the need to re-gather background information.
Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent template with about 20 custom skills to write long-form literature reviews. Sub-agents read thousands of papers, extracted key findings, stored them in a database, and then drafted the review section by section. Lecoq noted that a single review previously took his team up to two years; he now has roughly ten such reviews, many exceeding 100 pages. While this speed could accelerate synthesis, it also raises concerns about flooding the 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, used Claude Science for a glioma analysis. He said the tool completed the work in about a tenth of the time normally required, and his group verified the results manually, confirming they held up. This combination of speed and reliability is what researchers hope to see from AI-assisted discovery.
Commercial and regulatory context
The launch comes at a pivotal moment for Anthropic. The company has set ambitious revenue targets to justify its spending, and it is widely believed to be preparing for an initial public offering. Winning over paying customers in the scientific sector is a key part of that strategy. However, the timing is complicated by a tense standoff with the U.S. government over foreign access to its most powerful models. A product built for open scientific collaboration lands squarely in the middle of that geopolitical debate, potentially limiting its use by international partners.
Claude Science is available in beta for macOS and Linux users on Pro, Max, Team, and Enterprise plans. Academic and nonprofit labs can receive discounted seats. Additionally, Anthropic is funding up to 50 research projects with up to $30,000 in credits each, with applications open until July 15, 2026. The broader question remains: Can AI truly speed discovery, or will it simply produce more data and more papers? The labs now testing Claude Science will provide the first real-world answers.
Source:TNW | Anthropic News