NVIDIA BioNeMo Agent Toolkit Integrates with Anthropic's Claude Science for Accelerated Biology Workflows

Anthropic launched Claude Science, an AI workbench for scientific research, and integrated NVIDIA's BioNeMo Agent Toolkit to provide accelerated computational capabilities within the natural-language interface. BioNeMo packages NVIDIA-accelerated models (including Evo 2, Boltz-2, and OpenFold3), genomics tools (Parabricks), single-cell analysis (RAPIDS-singlecell), and cheminformatics (nvMolKit) as callable agents—allowing Claude to orchestrate multi-step lab workflows without manual configuration. The toolkit is open and framework-agnostic, available via GitHub and NVIDIA developer resources as of June 30.

The integration positions agentic AI as a runtime multiplier for scientific compute. A concrete example: a scientist can ask Claude to design inhibitors for a cancer target; Claude Science interprets the request, selects appropriate BioNeMo tools (conformer generation, clustering, validation), and returns results for iterative refinement—all while staying in natural language. Parabricks accelerates genomic analysis from hours to minutes; RAPIDS-singlecell compresses a 1.3M-cell workflow from 52 minutes to 25 seconds. NVIDIA reports 18 of the top 20 pharma companies already use BioNeMo, establishing an installed base for Claude Science adoption.

For infrastructure and ML engineers, this is a reference architecture for agent-as-adapter: frontier models (Claude) handle reasoning and planning; accelerated domain tools handle the compute-heavy steps. The pairing likely drives GPU utilization and token throughput in high-margin life-sciences deployments. Expect rapid uptake among biotech, pharma, and materials-science teams already running HPC workflows, as the barrier to agent-native acceleration drops to near zero.