Anthropic Launches Public Beta of AI Platform Claude Science for Research
# NVIDIA BioNeMo Speeds Up Anthropic Claude Science
## Anthropic has introduced the public beta of Claude Science, an AI platform that enhances scientific research.
Anthropic has unveiled the public beta of Claude Science, a groundbreaking AI workbench specifically designed to facilitate scientific research. This innovative platform allows scientists to interact directly with digital agents using natural language, allowing for the seamless execution of comprehensive research workflows. It integrates natively with the NVIDIA BioNeMo Agent Toolkit, giving users access to high-performance computing resources as easily callable skills within the Claude environment.
NVIDIA has developed what many recognize as the most extensive GPU-accelerated computing stack available today. This stack comprises physical hardware, software frameworks, operational libraries, scientific models, microservices, and domain-specific tools. Such a robust infrastructure enables researchers to conduct complex workflows and enhances their speed of iteration significantly.
Through this integration, scientists can import NVIDIA-accelerated models and computational libraries directly into their research environment. Notably, 18 of the world's top 20 pharmaceutical companies currently implement NVIDIA BioNeMo in their operations, signifying its strong foothold within the scientific community.
Claude Science translates natural language instructions into actionable results. This means researchers no longer need to manually configure predictive models or manage intricate software environments. By simply describing a research task—such as analyzing a genomic sequence or designing potential molecular binders—Claude Science interprets the request and orchestrates the execution with preconfigured, domain-specific agents.
These specialized agents comprehend established laboratory and computational protocols spanning genomics, proteomics, single-cell analysis, cheminformatics, and clinical research. The NVIDIA toolkit equips these agents with essential data context, linking each operational step to the relevant NVIDIA resources.
The toolkit packages NVIDIA-accelerated functions as specific callable programmatic skills, offering detailed information about each tool's function and data requirements. This setup empowers Claude Science to choose the appropriate computational tool, format valid data inputs, process tasks across NVIDIA's compute resources, and return outputs for human evaluation.
The integration creates a rapid feedback loop between human scientific inquiry and machine-accelerated processing. Scientists can assess generated outputs, refine their queries, and determine the next steps without diverting their attention from the core scientific questions.
For instance, producing enhanced inhibitors for known cancer targets showcases the practical implications of this system. A scientist may initiate the workflow by identifying a specific antigen mutation associated with cancer. Claude Science then collaborates with the BioNeMo Agent Toolkit to design multiple potential inhibitors targeting that mutation, thus expediting the complete pipeline of high-throughput inhibitor prediction, optimization, and validation.
The toolkit also provides access to advanced open models such as Evo 2, Boltz-2, and OpenFold3, empowering scientists with state-of-the-art biomolecular capabilities. Each model is tailored to ensure the autonomous agent has an appropriate scientific model for every phase of the research process.
AI agents in life sciences require specialized computational tools to plan and execute their tasks. A comprehensive workflow may involve fingerprinting an extensive compound library, clustering promising molecular candidates, and analyzing genomic context before recommending subsequent laboratory experiments.
An agent's efficiency is directly linked to the speed of its computational tools. With the NVIDIA BioNeMo Agent Toolkit, genomic analyses that previously took hours can now be completed in minutes, allowing agents to make real-time operational decisions incorporating complex genomic details.
Tools such as RAPIDS-singlecell, developed by scverse, compress extensive workflows from over 52 minutes down to just 25 seconds. This dramatic time reduction transforms single-cell analysis into an integral component of the agent’s reasoning loop, rather than a disjointed, delayed batch job. Similarly, the nvMolKit accelerates cheminformatics tasks like similarity searches by up to 3,000 times, delivering prompt results as agents traverse extensive chemical spaces.
To facilitate stable deployment of these advanced modeling workflows, NVIDIA packages its open biomolecular models into BioNeMo NIM microservices. These microservices are designed for enterprise use, functioning as high-performance inference endpoints.
Fully containerized, they feature a pre-tuned, accelerated software stack tailored for effective production environments. An autonomous agent can engage with these microservices through a single, consistent API to trigger remote deployments.
The NVIDIA BioNeMo Agent Toolkit is designed to be open and harness-agnostic, allowing scientific skills to operate uniformly across different agent frameworks and research platforms.
Engineering teams can access the toolkit and its associated skills via NVIDIA’s developer resources and GitHub repositories. During the current public beta phase, Anthropic invites feedback from researchers to inform future software integrations and identify additional domain specialists.
