Anthropic Launches Claude Science Research Workbench to Streamline Scientific Workflows

On Tuesday, Anthropic introduced Claude Science, a dedicated AI workbench designed to provide scientists with a unified environment for computational research, eliminating the need to constantly switch between databases, research pipelines, and specialised scientific tools.

The company emphasised that Claude Science is “not a new AI model and not a more capable model for biology.” Instead, it operates using the same Claude models already available to all users today, including Claude Opus 4.8, without providing exclusive model access or introducing additional capability restrictions.

The new workbench builds upon Anthropic’s Claude for Life Sciences initiative, introduced in October 2025, which enhanced the Claude chatbot for life science applications. Claude Science expands on that concept by providing researchers with a dedicated workspace specifically built to support scientific research workflows.

Announced during Anthropic’s AI for Science briefing on Tuesday, the launch reflects the company’s broader strategy of evolving beyond simply offering AI models. Much like Claude Code has become a workflow platform for software development, Anthropic aims for Claude Science to become the operating environment for scientific research. Rather than competing solely on model performance, the company is increasingly investing in industry-specific workflow products that could influence both its competitive positioning and pricing strategy.

The platform works by assigning a primary AI assistant that functions as a project manager for scientific research. It connects to more than 60 scientific databases and includes preconfigured toolkits covering disciplines such as genomics, protein structure analysis, and chemistry. That central assistant can create multiple specialised sub-assistants to divide complex research tasks, much like a project leader assigning work to subject-matter experts. Researchers can also deploy their own custom expert assistants tailored to individual research projects. Before any work is prepared for publication, a separate fact-checking AI reviews citations, calculations, and supporting material.

That verification layer has become increasingly important as AI-assisted scientific writing raises concerns about fabricated citations and unverifiable statistical information appearing in research papers. Anthropic notes, however, that the verification process still relies on the same underlying AI model rather than an independent external validation system.

According to Anthropic, Claude Science also improves the reproducibility of research. The workbench can generate outputs such as three-dimensional protein structures and chemical diagrams, along with the exact code used to create them. Every generated figure includes the original code, the software environment used during generation, a plain-language explanation describing how it was produced, and the complete message history associated with the task. Researchers can also edit figures using natural-language prompts, allowing the AI assistant to modify the underlying code automatically.

Another advantage highlighted by Anthropic is that Claude Science can operate within an organisation’s computing infrastructure rather than requiring research data to be processed on Anthropic’s cloud servers.

Several early users are already applying the platform to active research projects. Allen Institute neuroscientist Jérôme Lecoq has used Claude Science to build a multi-agent computational review pipeline. Meanwhile, Stephen Francis and his team at the UCSF Brain Tumour Centre employed the platform to significantly accelerate comprehensive germline analysis of glioma, reducing work that had previously taken much longer, while independently validating the resulting findings.

Claude Science enters a competitive market where OpenAI recently pursued a different strategy. In April, OpenAI introduced GPT-Rosalind, a specialised AI model specifically fine-tuned for biological reasoning tasks.

The distinction extends beyond the question of whether specialised AI models are necessary. It also reflects differing strategies regarding accessibility. GPT-Rosalind launched as a research preview available only to qualified U.S. enterprise customers following safety and eligibility reviews, with organisations including Amgen, Allen Institute, Moderna, Thermo Fisher, and Novo Nordisk receiving early access.

Google DeepMind has taken yet another approach. The company controls proprietary scientific foundation models, including AlphaFold and AlphaGenome, which competitors can access only as external tools. Its Gemini for Science platform also integrates those proprietary models alongside more than 30 life science databases within a single research environment.

As a result, three distinct approaches are now competing for the scientific AI market. Anthropic is pursuing broad subscription-based availability, OpenAI is targeting enterprise customers through controlled access, and Google is leveraging exclusive proprietary scientific models. The outcome could provide an early indication of how AI providers will compete across other specialised industries such as legal services, finance, and engineering.

Claude Science is currently available in beta for subscribers on Anthropic’s Pro, Max, Team, and Enterprise plans. The company also highlighted Novo Nordisk and the Allen Institute as customer examples, indicating that pharmaceutical and biomedical organisations are already engaging with multiple AI providers.

In addition, Anthropic announced support for up to 50 Claude Science research projects, offering up to $30,000 in usage credits per project. The company said it is seeking applications from graduate students and postdoctoral researchers working across scientific disciplines, with an initial emphasis on biomedical research. Applications remain open through July 15, 2026; award recipients will be notified by July 31; and selected projects will run from September 1 through December 1, 2026.