Anthropic launches Claude Science research workbench with NVIDIA BioNeMo toolkit

Anthropic announced Claude Science, an AI workbench for scientific research now in public beta, and NVIDIA simultaneously launched its BioNeMo Agent Toolkit that integrates directly with it. The toolkit turns NVIDIA's BioNeMo biomolecular models into callable skills for AI agents, backed by NIM microservices providing accelerated, containerized inference endpoints for drug discovery and life sciences.

The integration lets researchers access BioNeMo-powered workflows through Claude Science, running more sophisticated computational tasks and iterating faster using NVIDIA's full accelerated computing stack. Reported use cases include antibiotic design and vaccine prediction via text prompts, with partners like Basecamp Research contributing models — a striking demonstration of agentic AI operating on real biological data.

Competitively, this pushes Anthropic beyond coding and general assistance into specialized vertical science tooling, directly leveraging the protein-folding and computational-biology talent it has been hiring (including Nobel laureate John Jumper). It also deepens the Anthropic-NVIDIA partnership, which now spans model deployment on GB300 Blackwell Ultra in Azure and this science-specific toolkit.

The capability cuts both ways. Pharma and biotech researchers greeted the antibiotic-by-prompt claims with excitement, but the same power raises dual-use safety concerns — a text prompt designing antibiotics is uncomfortably close to a text prompt designing something harmful. Expect scrutiny of the guardrails Anthropic applies to Claude Science, especially given the recent government gating of its frontier models over security worries.