Inside OpenEvidence, Part 2: Rented Model, Rented Content, $15B Price Tag
This is the second part of an investigation into whether OpenEvidence, a medical AI company valued at $15B, has technology and content worth that price. The author finds the proprietary AI, content ownership, and accuracy claims all look weak under scrutiny.
- OpenEvidence doesn't own its content; publishers can pull their material at any time, which undercuts the moat a buyer would be paying for.
- Accuracy claims collapse outside exam settings: the company's near-perfect USMLE scores drop to about 27% on real-world clinical cases.
- Ad revenue projections ($400M to $500M in 2026) still put the $15B valuation at 30x to 37.5x a single year of sales, and the company's own CEO now pitches drug discovery as the real prize.
Part 2 of this investigation asks what a buyer would actually be getting if they paid $15B for OpenEvidence, the medical AI company that claims $300M in annual recurring revenue and raised at that valuation in September. The author's answer so far is that the case for a proprietary, defensible technology is weak. The piece looks at three things a buyer might be paying for: proprietary AI, proprietary content, and proven accuracy. It finds the content is not owned, since publishers can pull their material at any time, and the accuracy record is shaky. The author cites a result of 100% on the USMLE medical licensing exam but only 27% on real-world clinical cases, and a Nature Medicine paper that reportedly set off a public relations scramble at the company.
The article also covers four developments since Part 1. The Doximity lawsuit was dismissed with prejudice on October 16, removing a pending federal case that would normally worry an acquirer's diligence team. OpenEvidence announced a "landmark" collaboration with Johns Hopkins Medicine on October 13, with no financial terms or validation commitments attached. A JAMA usage study published October 15 tracked 271 million queries from roughly 950,000 clinicians, but three authors are tied to the company, the company supplied the data, and the study counts questions rather than correct answers. A Financial Times profile on October 20 reported unaudited figures of $100M in 2025 ad bookings, possibly including future commitments, and projected $400M to $500M in ad revenue for 2026. Even on that projection, the author points out, $15B is 30 to 37.5 times one year of ad sales, and CEO Daniel Nadler now frames drug discovery as the real prize.
The rest of the piece covers patient access restrictions, how the company and Nadler have handled critics, and whether OpenEvidence has a durable moat. The author's framing is skeptical throughout, and the article is explicitly opinion based on cited sources, with the company's non-responses to requests for comment noted. Readers who want the specifics on the accuracy studies and the critic-handling section should read the full piece, since the summary above can't carry the sourcing behind those claims.
Ask about this article
What does "rented model" mean in this context?
The article's title frames "rented model" against "rented content," and the section list points to the answer. I can give you the core meaning from the article's framing, and I'll pull in the public facts on the model side.
The phrase refers to OpenEvidence not owning the AI engine it runs on. The article's subheading treats the model as something the company licenses rather than builds. The section titled "What's Under the Hood (Still Undisclosed)" suggests the company hasn't said publicly which models power the product, so the claim is that a large part of the technology is borrowed from third parties.
Rented content is the parallel argument: section 2 says OpenEvidence "Owns None of Its Content. Publishers Can Pull It at Any Time." So the company's product depends on licensed or accessed material it doesn't control.
Put together, "rented model" means the core intelligence layer is a third-party foundation model, and "rented content" means the knowledge base is other publishers' material. Neither is owned, which undercuts the case that a $15B buyer is paying for proprietary AI or proprietary content. That's the logic the author uses in asking "what exactly are they buying?"
If you want to check the model side yourself, OpenEvidence's own site and its public statements are the place to look for which models it has named, and the article's section 1 is where the author says the company hasn't disclosed this.