OpenEvidence: The $12 Billion 'ChatGPT for Doctors' Most Physicians Already Use — and What It Means for FQHCs

In February 2025, an AI medical-search startup called OpenEvidence was worth $1 billion. By January 2026 it was worth $12 billion — and a reported 65% of U.S. physicians were using it. It is free to any clinician with an NPI, it is funded by pharmaceutical advertising, and it is built not on the open internet but on the New England Journal of Medicine, JAMA, and Cochrane. The New York Times broke its $6 billion fundraise. It is, quietly, one of the most-used pieces of AI in American medicine. For community health centers — where a single nurse practitioner may carry a panel no specialist will ever see — its rise is both the most promising and the most complicated AI story of the year.

The fastest valuation climb in health tech

The numbers are almost hard to believe. Sequoia led a $75 million round in February 2025 at a $1 billion valuation. By July, GV and Kleiner Perkins took it to $3.5 billion. In October, the New York Times' DealBook broke the news of a $200 million Series C at a $6 billion valuation, led by Google Ventures. Three months later, in January 2026, a $250 million Series D doubled the valuation again to $12 billion. That is roughly $700 million raised and a 12x valuation jump in about 11 months — with backers including Sequoia, Kleiner Perkins, GV, Blackstone, Nvidia, Coatue, and, tellingly, the Mayo Clinic.

The founder, Daniel Nadler, is not a first-timer: he built Kensho — an AI engine for Wall Street analysts — and sold it to S&P Global for about $550 million in 2018. His pitch for OpenEvidence is that physicians face 'the dark ages' of an information firehose, drowning in more medical literature than any human can read, and that the same kind of AI synthesis that worked for finance can work at the bedside.

How it reached most U.S. doctors

Adoption tracked the valuation. OpenEvidence says it went from ~25% of U.S. physicians at its Series A to 40% by mid-2025 to a reported 65% by April 2026, with roughly 760,000 registered physicians and monthly consultations climbing from 8.5 million to about 27 million. On March 10, 2026, it logged one million clinical consultations in a single day — the first time, the company says, any AI system did so with verified clinicians. One caveat worth stating plainly: these figures come from OpenEvidence itself and have not been independently audited.

Why it's free — and who actually pays

OpenEvidence costs clinicians nothing. It is paid for by pharmaceutical and medical-device advertising — and physician attention is some of the most expensive in advertising, with estimated CPMs of $70 to over $1,000 versus $5–$15 for consumer social media. The reason is uncomfortable: an ad can land at the precise instant a doctor is deciding what to prescribe. OpenEvidence says its advertising and clinical-answer systems are 'fully unconnected' and that advertisers 'cannot influence answers'; in May 2026 its CEO told NBC News the ad model may not be the company's long-term direction.

Built on NEJM and JAMA — but accuracy depends on the question

Here is what genuinely separates OpenEvidence from typing a question into ChatGPT: it trains only on licensed, peer-reviewed medical literature — content deals with NEJM Group, the JAMA Network's 13 journals, NCCN's oncology guidelines, the American College of Cardiology, the American College of Emergency Physicians, and Wiley plus the Cochrane reviews — roughly 35 million publications. Every answer is citation-linked. For a clinician, source-grounding is a real safety improvement over a general model that can confidently invent a reference.

But a headline benchmark can mislead. OpenEvidence became the first AI to score 100% on the USMLE in 2025 — yet an independent medRxiv pilot the same year found 34% accuracy on complex subspecialty board-exam questions with standard search, rising to 41% with its deeper 'DeepConsult' mode. A Mayo Clinic study of five academic primary-care cases found it accurate and evidence-aligned, but it mostly reinforced the physician's existing plan rather than changing it. Crucially, none of this testing used the patients an FQHC actually sees: uninsured, late-presenting, multilingual, with heavy social complexity. And it is not an FDA-cleared device.

The FQHC question: democratization, or the next equity gap?

This is where it gets interesting for community health. The single biggest barrier to new technology at FQHCs is cost — roughly 60% of community health centers, and 70% of rural ones, say so. OpenEvidence is free, runs in any browser with no EHR purchase, and is open to the nurse practitioners, physician assistants, and pharmacists who make up so much of the FQHC clinical workforce. For a solo clinician in a rural clinic with no specialist a phone call away, an instant, cited, evidence-based second opinion is exactly the kind of leverage that could narrow — not widen — the gap between safety-net and academic medicine.

And yet the warning signs are familiar. Every health system OpenEvidence has named as a partner — Sutter Health, Mount Sinai, Cedars-Sinai — is large and well-capitalized. No FQHC, community health center, or rural clinic is a documented deployment. The tool has never been validated on safety-net populations. The pharma-advertising model sits a click away from the formulary decision. And in California's Central Valley, where farmworker clinics serve some of the most medically underserved patients in the country, 20–30% of rural areas still lack reliable broadband — so even a free tool isn't free of friction. This is precisely the pattern health-equity researchers warn about: powerful technology reaching marginalized communities 'last, or not at all.'

OpenEvidence vs. ChatGPT for Clinicians — and the rules that (don't) apply

OpenEvidence is no longer alone. In April 2026, OpenAI launched a free, NPI-verified 'ChatGPT for Clinicians' tier — also free, also open to NPs and PAs, but without pharmaceutical advertising. The tradeoff is that it doesn't license the same top-tier journal databases OpenEvidence does. For an FQHC, the practical read is that the source-grounding and the funding model are the two axes that matter: OpenEvidence is the more rigorously sourced; ChatGPT for Clinicians removes the pharma conflict. Neither is a documented, validated safety-net tool yet.

What about the rules? California's AB 3030 — the generative-AI disclosure law — does not cover a clinician-facing reference tool: it applies to AI-generated patient communications that no licensed provider reviews, and a physician using OpenEvidence to inform their own judgment breaks that chain. SB 1120 governs payers' AI in utilization review, not bedside decision-support. So there is no disclosure trigger today. But that is not the same as no accountability: standard malpractice applies to any care decision, HRSA Operational Site Visits increasingly expect documented AI governance, and CHAI's new RUAIH responsible-AI certification (launched June 1, 2026) gives FQHC boards a concrete framework to point to. The right move for a community health center isn't to ban these tools or adopt them blindly — it's to write down who approved the tool, how staff are trained, and how errors get reported.

Frequently asked questions