Owkin Partners with NVIDIA, Rolls Out Multimodal Agent Platform Integrated with Claude
Owkin has entered its first collaboration with NVIDIA to advance the development of its biological large reasoning model, OwkinZero, using NVIDIA’s open AI infrastructure and reinforcement learning toolkit.
Announced today at JPM Healthcare Conference 2026 in San-Francisco, the partnership focuses on scaling agent-based biological modeling by integrating the Nemotron family of open models and NeMo RL, part of the NVIDIA NeMo framework for training large models with reinforcement learning.
Owkin was founded in 2016 by Thomas Clozel and Gilles Wainrib. Its architecture centers on biological large language models, multimodal patient data, and agent-based software designed to support reasoning across research and clinical settings. No financial terms for the NVIDIA collaboration were disclosed. The company has raised a total of $304.1 million to date across a $2.1 million seed round, a $72 million Series A, and a $230 million Series B.
The collaboration is positioned as a core milestone in Owkin’s roadmap toward Biological Artificial Super Intelligence (BASI)—a system of AI agents designed to reason across complex biological processes. Technical objectives include improving scalability, robustness, and model guardrails, with reinforcement learning identified as a key method for dynamic model customization and performance tuning at scale.
OwkinZero is the underlying engine for the company’s broader agentic architecture. It powers two core systems: Owkin K, an interoperable AI platform used by biopharma teams for research and discovery; and AI Scientist, an internal environment designed to automate hypothesis generation, target discovery, and diagnostic modeling. Both systems operate across multimodal biomedical data, enabling biological reasoning beyond traditional prediction-based approaches.
Concurrently with the partnership announcement, Owkin launched Pahology Explorer, its agentic infrastructure for biology designed to make its specialized AI agents deployable across research and clinical workflows. The infrastructure is built on curated multimodal patient data collected over a decade from more than 800 hospitals and is positioned as a foundational step toward the company’s stated goal of BASI.
Initial agents are intended to support tasks such as biomarker identification, multimodal data interpretation, and clinical trial decision support, with an emphasis on interoperability via APIs and Model Context Protocol compatibility. Owkin cited Pathology Explorer as an early example, reporting higher classification accuracy with fewer parameters compared with prior models, alongside reduced computational time.
Pathology Explorer has been embedded into Anthropic’s Claude for Healthcare and Life Sciences platform, also launched today at JPM2026, leveraging the Model Context Protocol (MCP) for seamless deployment. This enables Claude users, including pharmaceutical firms, hospitals, and research institutions, to access the spatially-aware pathology tools without altering their core workflows. The agent supports identification of cell and tissue types, tumor microenvironment analysis, and biomarker extraction from digitized slides.
Throughout 2025, a string of announcements expanded Owkin’s agentic AI footprint. In October, Owkin launched K Pro, a natural language interface for researchers and pharma teams to explore multimodal biological data. Earlier in the year, it began deploying Owkin K at the Medical University of Vienna, starting with use in advanced bladder cancer.
In November, Owkin unveiled a pan-European initiative with Gustave Roussy and Charité to build cross-border AI infrastructure for biology, designed to make biomedical data research-ready and support European digital sovereignty. This January, Servier began evaluating patient subgroups identified by Owkin’s models in Phase 2 trial data.
Topic: Next-Gen Tools
