KK Pro by Owkin

Two of the world’s largest drugmakers signed up for K Pro within a month of each other in 2026: AstraZeneca with a three-year license in May and Sanofi with a five-year license in June. In both deals Owkin, a biotech AI company with offices in Paris, New York and Geneva, builds custom agents on K Pro that carry out drug research, development and competitive intelligence work inside the customer’s own systems. The agents draw on curated patient data from more than 800 hospitals.

K Pro is best for large pharma companies willing to enter a multi-year partnership. It is built for partnerships rather than quick self-serve trials: pricing comes through negotiation, there is no self-serve plan and Owkin leads the development of each customer’s agents. We rate K Pro 7.4/10 on our scale. We have not used the product; this review is based on Owkin’s announcements and news coverage of its deals.

Paying Customers, Data and Company Benchmarks

Owkin’s best evidence is who is paying for it and the data behind its models, plus a set of company-reported benchmarks.

The two pharma licenses are the strongest signal. AstraZeneca and Sanofi have their own AI teams, and both chose multi-year terms with Owkin building the agents. Sanofi also has a long history with the company, starting with a €90 million partnership in 2021, so its five-year commitment reflects years of working together rather than a first trial.

The benchmarks are best read as starting points. The 23.7% accuracy gain and the 20x speedup are Owkin’s own figures, averaged across tasks Owkin chose. They are useful for deciding what to pilot, not for predicting results on your data.

The partnerships with Anthropic and NVIDIA point in two directions. Distribution through Claude puts a specialized Owkin agent in front of researchers who never sign a K Pro license, while NVIDIA tooling supports the OwkinZero model that sits underneath the enterprise agents.

Orchestrated Skills and Tools for Pharma Research

Owkin describes K Pro as an AI scientist for pharmaceutical research and strategic decisions. It orchestrates AI skills and tools on top of three assets: Owkin’s multimodal patient data network, its biological AI models and about a decade of work with pharma partners.

Custom Agents for Each Customer

In the AstraZeneca and Sanofi deals, Owkin leads end-to-end development of agents built for the customer. AstraZeneca’s agents are integrated into its IT systems and decision workflows, with competitive intelligence as the first focus, aiming to cut reliance on manual analysis while meeting the company’s governance, security and enterprise standards. Sanofi’s agents are meant to perform complex research and development tasks autonomously and to complement the agentic AI tools Sanofi already runs.

Specialized Agents Through an API

In January 2026 Owkin released a series of autonomous agents for drug discovery and research, delivered through an API that supports the Model Context Protocol. The first, Pathology Explorer, is available through Anthropic’s Claude for life sciences offering. Built on Owkin’s Histo Interpretability Prediction Engine model, it identifies cell types, their locations and biomarkers from digital pathology images alone. Owkin says the agents improve through refinement with new patient data and lab-in-the-loop testing.

Models and Data

• Patient data: a curated multimodal dataset from more than 800 hospitals, which Owkin says grounds its agents in clinical insight from real patients.
• OwkinZero: Owkin’s biological reasoning model, which it is refining with NVIDIA NeMo and the cuCIM image processing library.
• MOSAIC: a spatial omics dataset in oncology from academic partners including Charité, the University of Pittsburgh, Gustave Roussy, CHUV and Uniklinikum Erlangen.

Benchmarks Owkin Has Published

Owkin tested K Navigator against a subset of the public LAB-Bench benchmark and reported performance on par with frontier models, including OpenAI’s o3 reasoning model, on most tasks, with an edge on a task that asks models to identify genes in cancer-related gene sets. For Pathology Explorer, Owkin reported 23.7% better classification accuracy on its chosen benchmarks while using five times fewer parameters than comparable agents, and computation cut from weeks to hours.

What Customers and Owkin Say

Owkin co-founder and CEO Thomas Clozel has said the future of the pharmaceutical industry is agentic, and described the Sanofi deal as a move toward AI embedded in a company’s own workflows. Sanofi’s chief digital officer, Emmanuel Frenehard, tied the deal to helping Sanofi teams work with more speed, depth and confidence. Early K Navigator testers at University College London and Institut Curie praised its ease of use and the independence it gave researchers to explore patient data. These are selected endorsements, so weigh them alongside your own pilot.

K Navigator for Researchers

K Navigator, launched in May 2025, is the researcher-facing co-pilot. It lets scientists query and visualize data in natural language, review literature across 26.5 million articles, draw on 19 biomedical databases and explore a 60-patient subset of MOSAIC at single-cell resolution. It opened as a free six-month beta for a limited group, with a waitlist.

Customers Decide What Owkin’s Agents May Do

Owkin builds the agents, but the customer decides what they are allowed to do and acts on the results. Competitive intelligence briefings, target hypotheses and pathology readings still need expert review before they shape portfolio or clinical decisions.

Only Deal Lengths Are Public

Owkin does not publish K Pro prices. Public information is limited to deal length.

Time and Commitment Math for a Multi-Year License

With no published prices, the useful math is about time and commitment. Volumes below are our own illustrative assumptions applied to Owkin’s published figures.

• Contract length: a three-year license like AstraZeneca’s covers 36 months and a five-year license like Sanofi’s covers 60, so total cost is set by a long commitment rather than monthly usage.
• Target characterization: if annotating a drug target takes an analyst 20 hours by hand, Owkin’s 75% faster figure would bring it to 5 hours; across 40 targets a year that is 600 hours saved.
• Spatial analysis: producing a spatial transcriptomics plot in about one minute instead of about three days changes how many exploratory plots a team will try, which matters more than the hours alone.

Other costs include internal IT integration work, governance review for patient-data-based outputs and the staff time to work alongside Owkin’s developers during agent builds.

K Pro Compared With IQVIA.ai, Causaly Agentic Research and Kosmos

Enterprise research agents differ mostly in the data they bring. Owkin brings hospital patient data, IQVIA brings commercial and clinical data, Causaly brings a biomedical knowledge base and Kosmos works on whatever data you upload.

K Pro is the closest thing in this group to an outsourced agent team. Kosmos is the self-serve opposite, and IQVIA.ai sits between them with a catalog plus services. See more on our life sciences agents page.

Large Pharma Research Groups Ready to Partner

• Large pharma R&D organizations: the AstraZeneca and Sanofi deals show K Pro is built for this scale and this kind of partnership.
• Competitive intelligence teams: AstraZeneca’s first agents target faster, data-driven competitive insight.
• Oncology and pathology groups: Pathology Explorer reads pathology images and the MOSAIC spatial data comes from oncology research.
• Teams already on Claude: Pathology Explorer is available through Anthropic’s Claude for life sciences offering.

Small Biotechs and Academic Labs

• Small biotechs and startups: multi-year licenses and Owkin-led builds are sized for large companies.
• Academic labs wanting instant access: K Navigator opened as a limited beta with a waitlist; a self-serve option such as Kosmos or Elicit is easier to start.
• Buyers who need published prices: all K Pro terms are negotiated.
• Commercial and field teams: agents inside CRM, such as Veeva AI Agents, fit sales work better.

Benchmark Scope and Ownership of Custom Builds

• Vendor benchmarks: the 23.7% accuracy gain and 20x speedup are Owkin figures on tasks it selected.
• Ownership of custom builds: when Owkin builds your agents, ask who owns the agents, prompts and fine-tuned models if the license ends.
• Patient data governance: outputs grounded in hospital data need clear rules on provenance, consent and reuse.
• Company changes: Owkin spun its diagnostics division out as Waiv and says it is focusing on biology, so check which teams support which products.
• Long commitments: three- and five-year terms make an early pilot and clear exit terms important.
• Limited researcher access: K Navigator opened to a small group with a waitlist, so individual scientists may not be able to test the technology before their company signs.

K Navigator’s 2025 Launch as a Research Co-Pilot

• On May 6, 2025, Owkin launched K Navigator, a researcher co-pilot it said speeds up biomedical research tasks by 20x on average.
• On January 12, 2026, SiliconANGLE reported Owkin’s first autonomous agents, including Pathology Explorer through Anthropic’s Claude, and an NVIDIA partnership for OwkinZero.
• On May 13, 2026, Owkin signed a three-year K Pro license with AstraZeneca to build agents, starting with competitive intelligence.
• On June 5, 2026, Owkin signed a five-year K Pro license with Sanofi to co-develop biopharma agents.

A Shortlist Candidate for Multi-Year Pharma Partners

For large pharma companies ready for a multi-year partnership, K Pro deserves a place on the shortlist. It scores 7.4/10 on our scale, a figure that reflects negotiated pricing and an involved setup model rather than doubts about the science. Two top drugmakers signing multi-year licenses within weeks is strong validation, and patient data from more than 800 hospitals grounds Owkin’s agents in real clinical data.

For smaller teams, the model is a less natural fit. There is no self-serve route, and value depends on Owkin’s team building agents around your workflows. If you are evaluating it, start with one bounded use case such as competitive intelligence, set baseline times before the build and negotiate ownership of the custom agents up front.

K Pro Versus K Navigator

How Much Does K Pro Cost?

Owkin does not publish pricing. Known customers sign multi-year licenses, three years for AstraZeneca and five years for Sanofi, and the financial terms were not disclosed.

What Is the Difference Between K Pro and K Navigator?

K Pro is Owkin’s AI scientist platform for pharma, with custom agents built by Owkin. K Navigator is the researcher co-pilot launched in May 2025 as a free six-month beta for a limited group.

Can I Use Owkin Agents in Claude?

Yes, for Pathology Explorer. Owkin made it available through Anthropic’s Claude for life sciences offering, and its agents are delivered through an API that supports the Model Context Protocol.

What Data Do Owkin Agents Use?

Owkin says its agents draw on curated multimodal patient data from more than 800 hospitals. K Navigator also uses 26.5 million articles, 19 biomedical databases and spatial omics data from the MOSAIC project.

How Does K Pro Compare With Kosmos?

Kosmos is self-serve, with free academic credits and a $200 monthly plan, and runs on data you upload. K Pro is a multi-year enterprise license where Owkin builds agents on its own patient data for each customer.