Anthropic pivots to scientific workflow with new Claude workbench

The platform functions as a project management hub, allowing a primary AI assistant to delegate tasks to specialized sub-agents. These tools handle complex domains including genomics, chemistry, and protein structure analysis. To address the persistent issue of AI-generated hallucinations, the workbench includes a dedicated fact-checking layer that verifies citations and calculations before research reaches the publication stage. While the system relies on existing Claude models, it emphasizes reproducibility by linking every generated figure—such as 3D protein models—directly to the specific code, environment, and message history that produced it.

This release marks a strategic shift for Anthropic, which is increasingly focused on owning the operational layer for specific industries rather than competing solely on raw model capability. By allowing labs to run processes on their own infrastructure and providing tools for plain-language figure editing, the company is positioning itself against rivals like OpenAI and Google DeepMind. While OpenAI has opted for gated, enterprise-only biological models like GPT-Rosalind, and DeepMind leverages proprietary foundational models like AlphaFold, Anthropic is targeting broader accessibility. The beta is currently available to users on Pro, Max, Team, and Enterprise subscription tiers, with the company offering up to $30,000 in credits to support academic research projects through late 2026.