The biological reasoning model
powering Owkin K
What is OwkinZero?
OwkinZero is a model purpose-built for biological discovery.
Trained on 300,000+ verified Q&A pairs across eight key biomedical tasks, it draws on proprietary patient data, spatial and bulk transcriptomics, and Owkin’s domain expertise.
Why we developed it
Generic AI models struggle to deliver actionable, biologically-grounded insights for drug development.
Biopharma needs decision-grade answers, not just data.
We developed OwkinZero to power K Pro, our agentic AI decision-making solution for biopharma.
How OwkinZero works
OwkinZero's performance
OwkinZero consistently outperforms generalist LLMs on several specialist biological tasks
Smaller fine tuned models are able to outperform frontier models on specialist tasks.
Fine tuning on many contexts is better than one.
Generalization to unseen tasks is still difficult.
How are OwkinZero models trained?
Raw data is transformed into natural language
- 🟢 CC Stroma
- 🔵 CC Tumor Islet Edge
- 🟠 CC Tumor Islet
QUESTION: “Which gene is downregulated in tumor islets versus stroma in Bladder urothelial carcinoma?”
OPTIONS: {“A”: “AFAP1”, “B”: “BPIFA1”}
ANSWER: “B”
Q&A datasets covering several aspects of biology
Q&A datasets are used for Reinforcement Learning
Eight comprehensive benchmark datasets targeting critical challenges in drug discovery, including target druggability, modality suitability, and drug perturbation effects
Post-train OwkinZero models using GRPO from Qwen3 models
How OwkinZero powers K Pro
Complex biological questions are routed to its specialized engine, which integrates multiomics, literature, and clinical data for unified, evidence-backed answers.