Anthropic launches Claude Science workbench for computational research
Anthropic’s Claude Science workbench centralizes computational research into one environment, linking 60+ databases, toolkits and reproducible workflows (beta).
Anthropic on June 30, 2026 launched Claude Science, a dedicated workbench that consolidates databases, pipelines and tools for computational life sciences research. Claude Science is not a new model; it runs the same Claude models available to customers today and is designed to give scientists a single environment for project work. The company positions the product as an operating layer for scientific workflows, aiming to reduce the time researchers spend switching between fragmented systems.
Workbench purpose and model access
Anthropic describes Claude Science as a unified workspace rather than a bespoke biology model, and it runs existing Claude models such as Claude Opus 4.8 with no special gating. The company said the workbench extends capabilities introduced with Claude for Life Sciences last October, packaging those capabilities into a project-oriented environment. By keeping the underlying models unchanged, Anthropic emphasizes that Claude Science focuses on workflow integration and tooling rather than a shift in model capability.
How the assistant coordinates research tasks
At the center of Claude Science is a principal assistant that acts like a project manager for teams, connecting to more than 60 scientific databases and offering pre-built toolkits for genomics, protein structure and chemistry. That main assistant can spawn sub-assistants to divide tasks, hand off work to user-created expert assistants, and orchestrate multi-step pipelines. A separate automated fact-checker reviews citations and calculations before outputs are finalized, a step Anthropic says is meant to reduce common errors in AI-assisted manuscripts.
Reproducibility through provenance and editable outputs
Claude Science captures the provenance for generated figures and code by saving the exact code, execution environment and message history that produced each result. Figures such as 3D protein structures or chemistry diagrams are paired with the runnable code and a plain-language explanation of their creation. Users can request edits in natural language, prompting the agent to update the underlying code so visual outputs and scripts remain reproducible and auditable.
On-premise deployment and data control options
To address data sensitivity, Claude Science can run on a laboratory’s own infrastructure rather than routing research data through Anthropic’s servers. That flexibility is intended to help organizations maintain control over confidential datasets and comply with internal governance or regulatory constraints. Anthropic presents on-premise execution as a time-saving and privacy-preserving option for teams with strict data-handling requirements.
Early users and demonstrated workflows
Anthropic cited early adopters who say the workbench accelerates development timelines. A principal scientist at Gladstone Institutes reportedly built a genome browser from scratch in days, while a researcher at the Allen Institute used the platform to create a multi-agent computational review pipeline that the company says dramatically reduced manual labor. Anthropic also named Novo Nordisk and the Allen Institute as case studies, signaling pharma and neuroscience labs among its initial customer base.
Competitive landscape and differing strategies
Claude Science arrives alongside competing efforts that take different technical approaches to biological research support. OpenAI released GPT‑Rosalind in April 2026 as a biology-focused model available via a gated research preview for vetted enterprise partners. By contrast, Google DeepMind offers foundational science models such as AlphaFold and AlphaGenome and bundles those capabilities in Gemini for Science, which integrates proprietary models with dozens of life-science databases. Anthropic’s strategy is to deliver a vertical, workflow-level product built on its general-purpose Claude models rather than a specialized biology model.
Beta terms and research program details
Claude Science is available in beta to users on Anthropic’s Pro, Max, Team and Enterprise plans, and the company is inviting researchers to apply for supported projects. Anthropic will fund up to 50 projects with credits of up to $30,000 each, targeting postdoctoral and graduate work that explores biomedical research domains. Applications are open through July 15, 2026, award notifications will be issued by July 31, and selected projects are scheduled to run from September 1 to December 1, 2026.
Anthropic’s Claude Science workbench aims to shift how institutions build computational research pipelines by embedding tooling, provenance and multi-agent coordination in a single environment. Whether the approach of packaging existing Claude models into a workflow platform will reshape lab practices depends on adoption among large research organizations and how the product compares in speed, safety and interoperability to offerings from other major AI developers.
