Quick answer: Claude Science is Anthropic’s new beta AI workbench for scientists. It brings scientific databases, analysis tools, code execution, artifact provenance, and compute management into one research environment. The most useful angle for labs is not “another chatbot” — it is a reproducible workflow layer for literature review, figures, notebooks, manuscripts, and HPC-style jobs.
| Product | Claude Science beta |
| Official launch | Announced by Anthropic on June 30, 2026 |
| Who it is for | Scientists, research labs, life-science teams, computational researchers, and teams that need auditable AI-assisted analysis |
| Access | Anthropic says beta access is for Claude Pro, Max, Team, and Enterprise users |
| Platforms | macOS, Linux, remote machines over SSH, HPC login nodes, and related compute setups |
| Key promise | Research outputs with traceable code, environment, conversation history, and evidence checks |
Claude Science checklist: should your lab try it?
- You handle reproducible analysis: figures, tables, notebooks, or manuscripts need to be traceable back to code and data.
- You switch between many tools: PubMed/literature search, Python, R, shell, Jupyter, HPC, databases, and visualization tools.
- You need review before publishing: Claude Science includes a background reviewer that Anthropic says can flag unsupported claims, citation issues, untraceable numbers, and figures that do not match code.
- Your data should stay where it is: Anthropic says large or sensitive datasets do not have to leave existing systems; only needed context is sent to Claude.
- You use scientific compute: Claude Science can draft plans, request permission before accessing resources, and submit jobs to existing lab infrastructure.
What Claude Science actually does
According to Anthropic, Claude Science is an app that integrates common scientific tools and packages, produces auditable artifacts, and gives researchers flexible access to computing resources. In practical terms, it is meant to reduce the back-and-forth between literature search, analysis notebooks, data files, visualization code, manuscript drafts, and compute environments.
Anthropic describes Claude Science as a workbench where researchers can analyze literature, run multi-step projects, generate and refine figures, draft manuscript sections, and manage compute while keeping an auditable record of how outputs were produced.
Best use cases
| Use case | How Claude Science helps | Human review needed? |
|---|---|---|
| Literature review | Searches and summarizes scientific sources, with citations to verify | Yes — verify every important claim |
| Figure generation | Creates and edits figures from code and lets users request changes in natural language | Yes — inspect code and data |
| Genomics / single-cell / proteomics workflows | Uses curated skills and connectors for domain-specific pipelines | Yes — confirm parameters and assumptions |
| HPC or GPU jobs | Drafts compute plans and can submit jobs to existing infrastructure after review | Absolutely — review before execution |
| Manuscript drafting | Writes results sections close to the analysis that produced them | Yes — scientific authors remain responsible |
Access, platforms, and credits
- Beta availability: Anthropic says Claude Science is available in beta for Claude Pro, Max, Team, and Enterprise users.
- Admin requirement: Team and Enterprise customers may need an administrator to enable Claude Science.
- Supported environments: Anthropic lists macOS, Linux, remote machines through SSH, and HPC login nodes among the supported ways to work.
- Research credits: Anthropic says it will support up to 50 Claude Science AI for Science projects with up to $30,000 in credits. Check the official page before applying because eligibility and deadlines can change.
- Research lab plan: Anthropic also has a Claude Team plan page for academic and nonprofit research labs.
Prompt templates for researchers
Use these as starting templates and adapt them to your actual dataset, lab policies, and review process.
Literature map prompt:
Search recent papers on [topic]. Build a table with: paper, method, dataset, key finding, limitation, and citation. Flag claims that need manual verification.Figure audit prompt:
Inspect the code and data used for this figure. Explain exactly how it was generated, list assumptions, and flag any mismatch between the figure and the underlying analysis.HPC job review prompt:
Draft a compute plan for this analysis before running anything. Include required inputs, estimated compute, software environment, expected outputs, failure points, and a rollback/checkpoint plan.Important limitations
- Claude Science is in beta, so workflows, access, pricing, and limits may change.
- AI-generated scientific outputs still need expert review, especially for clinical, safety, regulatory, or publication decisions.
- Do not upload confidential, patient, proprietary, or regulated data unless your institution has approved the workflow and account configuration.
- For grant applications or credits, rely on Anthropic’s official pages rather than third-party summaries.
Sources
- Anthropic: Claude Science, an AI workbench for scientists
- Claude Science beta product page
- Claude Team plan for research labs
FAQ
Is Claude Science a new model?
No. Based on Anthropic’s description, Claude Science is a beta workbench app for scientific workflows, not simply a new model name.
Can Claude Science run jobs on a lab cluster?
Anthropic says Claude Science can work with environments such as remote machines over SSH and HPC login nodes, and can help submit jobs to existing lab infrastructure after user review.
Should researchers trust Claude Science outputs automatically?
No. Treat it as an assistant for analysis and documentation. Researchers should verify citations, code, calculations, datasets, and conclusions before using outputs in publication, clinical, regulatory, or commercial work.