Anthropic launched Claude Science, an AI workbench designed for computational research that runs on existing Claude models rather than a specialized one. It connects to over 60 scientific databases and uses a multi-agent architecture where a main assistant acts as project manager, spawning sub-assistants for specific tasks and a separate fact-checker for citations and calculations. The platform supports reproducibility by bundling generated figures with the code and environment that produced them. Early users at Gladstone Institutes and Allen Institute report significant time savings. Claude Science is positioned as a vertical, workflow-level product — similar to how Claude Code targets software development — contrasting with OpenAI's GPT-Rosalind (a fine-tuned biology model) and Google DeepMind's approach of owning foundational science models like AlphaFold. It's available in beta to Pro, Max, Team, and Enterprise subscribers.
3.5K Impressions2 Comments