Anthropic just gave scientists their own dedicated home inside the Claude ecosystem — and it has nothing to do with launching a smarter model.
What Is Claude Science?
On June 30, 2026, at its “The Briefing: AI for Science” event, Anthropic launched Claude Science, a new application built specifically for researchers. It is best described as a workbench rather than a chatbot: a single workspace where scientists can pull together literature, lab data, code, and compute resources that today live scattered across dozens of disconnected databases and tools.
Anthropic is explicit that this is not a new or more capable model for biology. Claude Science runs on the same Claude models already available to everyone, including Claude Opus 4.8, with no special access tier or gating. The innovation here is entirely at the workflow layer — the same strategy that turned Claude Code into the default operating layer for software development is now being pointed at laboratories.
Researchers typically lose enormous amounts of time switching between tools like PubMed, Jupyter notebooks, R, and cluster terminals, each with its own schema and quirks. Claude Science folds all of that into one environment where a generalist coordinating agent can analyse literature, execute multistep research, and progressively refine figures and manuscripts until they are publication-ready. Every output comes with an auditable trail back to the exact code and environment that produced it, so results can be validated and reproduced months later.
How It Actually Works
A workbench, not a website
Claude Science behaves more like a Jupyter Notebook than a typical AI app. It can be accessed locally on macOS or Linux, or on a remote machine over SSH or an HPC login node — wherever a researcher already does their work. This matters for labs handling sensitive or massive datasets, since the data never has to leave the systems it already lives on; only the relevant context for each step is sent to Claude.
You can access Claude Science from this link.
60+ skills, pre-built for science
Out of the box, Claude Science comes pre-configured with more than 60 curated skills and connectors spanning genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. It connects natively to major scientific databases such as UniProt, PDB, Ensembl, Reactome, ClinVar, and ChEMBL, and it taps NVIDIA’s BioNeMo Agent Toolkit to reach life-science models including Evo 2, Boltz-2, and OpenFold3. Ask a question in plain language, and specialist sub-agents query and synthesise across these sources automatically.
Built-in fact-checking
A dedicated reviewer agent runs alongside the main workflow, checking citations and calculations and flagging anything that looks off — incorrect references, untraceable numbers, or figures that don’t match their underlying code — and correcting errors as it goes. Sessions also hold context in memory, so massive datasets only need to be loaded once, and researchers can fork a session at any point to test a different approach without losing their original thread.
Compute, handled automatically
Large jobs — folding a protein or running a genomics pipeline across a huge dataset — normally mean researchers themselves have to configure compute jobs, wait, and babysit them. Claude Science drafts a plan, asks before reaching for new resources, and then submits the job to whatever infrastructure the lab already uses, whether that’s an in-house HPC cluster or on-demand GPUs through a Modal account, scaling from a single GPU to hundreds as needed.
Real-World Use Cases Already in Motion
- Gladstone Institutes: principal scientist Sean Whalen used Claude Science to build a genome browser from scratch in a matter of days.
- Allen Institute: neuroscientist Jérôme Lecoq built a multi-agent “computational review” pipeline of roughly 20 custom skills that reads thousands of papers and drafts long-form literature reviews — work that used to take his team up to two years now produces around ten 100-plus-page reviews, with citations checked by reviewer agents.
- Manifold Bio: used Claude Science end-to-end to nominate drug targets for tissue-specific medicines, ranking candidates against the company’s own proprietary research criteria.
- UCSF Brain Tumor Center: epidemiologist Stephen Francis used it to accelerate glioma research, cutting comprehensive germline genetic workups to roughly one-tenth of the time they previously took, with results independently validated by his lab.
Anthropic has also named Novo Nordisk as a customer case study, signalling that large pharmaceutical organisations are already experimenting with the tool alongside other AI vendors.
Who Should Use Claude Science?
Despite the broad “Science” branding, the current release is squarely aimed at life sciences and biomedical research:
- Academic researchers and postdoctoral or graduate students working in genomics, single-cell biology, proteomics, structural biology, or cheminformatics.
- Pharmaceutical and biotech R&D teams running drug discovery, target identification, or clinical research workflows.
- Computational biologists and bioinformaticians who currently juggle multiple specialised databases and pipelines.
- Research labs at universities and nonprofits, for whom Anthropic now offers a discounted Team plan specifically for academic and nonprofit science labs.
Notably, Anthropic says you don’t have to be a scientist to use it — it’s available to anyone on a Pro, Max, Team, or Enterprise Claude subscription, though Team and Enterprise admins need to switch it on first. The naming choice (“Science,” not “Biology” or “Life Sciences”) strongly hints that Anthropic intends to expand into physics, materials science, and other research domains over time.
Why This Matters: The Bigger Picture
Claude Science arrives as Anthropic pushes hard to diversify revenue ahead of an anticipated IPO, having reportedly reached $42 billion in annualised sales and a $965 billion valuation off a recent funding round. Alongside Claude Code, Claude Cowork, Claude Design, and Claude Security, Science is the latest bet that Anthropic’s growth will come from owning workflow-level products in specific industries rather than competing purely on raw model capability.
It also lands squarely in a heating-up race. OpenAI released GPT-Rosalind in April — a model specifically fine-tuned for biological reasoning — but gated it as a research preview limited to qualified U.S. enterprise customers behind a safety review. Google, meanwhile, launched “Gemini for Science” at I/O in May, also pitched as a desktop scientific workbench spanning 30-plus life-science databases plus an “AI co-scientist” hypothesis engine. Anthropic’s point of difference is openness of access: Claude Science is in beta for any paid subscriber, no qualification process required, while rivals lean toward specialised models with tighter gating.
To support adoption beyond the big labs, Anthropic is funding up to 50 “AI for Science” projects with up to $30,000 in credits each (plus up to $2,000 in Modal compute for select projects), prioritising postdoctoral and graduate research that explores new scientific frontiers, with an early focus on biomedical fields. Applications are open through July 15, 2026, with awards announced by July 31 and project work running from September through December 2026.
Key Takeaways
- Claude Science is a workbench app, not a new model — it runs on existing Claude models like Opus 4.8.
- It unifies literature review, data analysis, compute management, and manuscript drafting into one auditable workspace.
- It ships pre-configured with 60+ skills and connectors for genomics, proteomics, structural biology, and cheminformatics.
- Available in beta now for Claude Pro, Max, Team, and Enterprise users on macOS and Linux.
- Early users include Gladstone Institutes, the Allen Institute, Manifold Bio, UCSF, and Novo Nordisk.
- It competes directly with OpenAI’s GPT-Rosalind and Google’s Gemini for Science, but with broader, less gated access.
Frequently Asked Questions
Is Claude Science a new AI model?
No. Anthropic has been explicit that Claude Science runs on its existing Claude models, including Opus 4.8, with no special or more capable model underneath. The innovation is the workflow and tooling layer built around the model.
Who can access Claude Science?
It’s available in beta to anyone on a Claude Pro, Max, Team, or Enterprise plan, on macOS or Linux. Team and Enterprise admins need to enable it for their organisation first.
How is Claude Science different from Claude Code or Claude Cowork?
Where Claude Code is built around software development workflows, Claude Science is purpose-built for scientific research — pre-loaded with domain-specific databases, connectors, and skills for fields like genomics and structural biology, and capable of natively rendering protein structures, genome tracks, and chemical diagrams.
How does Claude Science compare to OpenAI’s GPT-Rosalind and Google’s Gemini for Science?
GPT-Rosalind is a specialised model fine-tuned for biological reasoning, currently gated to qualified enterprise customers under a safety review. Gemini for Science is a similarly positioned desktop workbench. Claude Science differentiates itself by running on general-purpose Claude models with open beta access for any paid subscriber, rather than a specialised, tightly gated model.
Sethu Ram is a search strategist with 16+ years of experience in international SEO across EMEA, APAC, MENA, and North America. He runs WorthView as a live lab for GEO and AI search experimentation, covering the intersection of generative AI, search evolution, and what it means for publishers navigating the post-blue-link web. He is also the founder of MoneyHulk, a personal finance publication for Indian audiences.
One thought on “Claude Science: Anthropic’s New AI Workbench Wants to Be the Claude Code of Lab Research”