What Is OpenEvidence? The Clinical AI Used by Roughly 40% of U.S. Physicians (2026 Guide)

Hello, this is Ryuta Hamamoto from TIMEWELL.

People often describe OpenEvidence as “ChatGPT for doctors.” That label is catchy, and it is incomplete. The product’s real design choice is narrower and more useful: ground answers in peer-reviewed medical literature, always show the citations, and limit deeper clinical use to verified clinicians. That is a different evolutionary path from general-purpose chatbots, and it has lessons that reach well beyond medicine, especially for enterprise AI that has to handle confidential knowledge.

In my consulting work I still get the same two questions almost every week. Is it safe to drop internal documents into a public LLM? How do we design around hallucination without pretending the problem is solved? When you take OpenEvidence apart carefully, one workable answer starts to show up. I have updated this piece several times through 2026 as revenue, valuation, partnerships, and EHR deals kept moving. Below is the picture I can confirm as of this rewrite: what the product is, how adoption and funding scaled, how the citation system works, what U.S. clinical safety questions actually look like, how non-U.S. clinicians are using it, and what enterprise teams can borrow. If you want a quick read on AI rollout economics before you go further, our free AI ROI calculator gives a rough estimate in a few minutes.

What OpenEvidence is: clinical AI used by roughly 40% of U.S. physicians

As a legal entity, OpenEvidence is a U.S. medical AI company incorporated in September 2021¹. Full commercial rollout ramped in 2023 after work with the Mayo Clinic Platform accelerator, where the team appears to have tightened medical information structure and retrieval against real clinical feedback¹. Holding the timeline as “company in 2021, full product push from 2023” matches the growth curve that followed.

Founder and CEO Daniel Nadler previously built Kensho, an AI company focused on economic data, and later sold it to S&P Global for roughly $550 million¹. Why medicine for a second company? Contrary Research’s account ties the choice to personal history. Nadler lost his grandfather to a medical error. Co-founder Zachary Ziegler, who researched machine learning at Harvard, had a brother-in-law who went through leukemia treatment¹. The company did not spin out of a single health system. That independence later helped it partner across publishers and societies rather than locking to one system’s stack.

Adoption speed is the headline number. Sacra’s tracking for the first half of 2026 puts daily use at roughly 40% of U.S. physicians, more than 10,000 institutions, and about 65,000 new registrations a month². OpenEvidence’s own August 15, 2025 release said it was already providing evidence-based clinical decision support to more than 40% of U.S. physicians³. This is no longer a pilot tool. It sits closer to a new default layer at the front of U.S. clinical information seeking.

Business metrics moved just as fast. Sacra’s May 2026 update puts full-year 2025 revenue at about $150 million, roughly 18 times the $7.9 million reported for 2024, or about +1,803% year over year. Gross margin is reported around 90%, and ARPU around $124². The important detail is that this scale was built without charging physicians for core clinical use.

Metric	Value	Notes
Daily use among U.S. physicians	~40%	Used at 10,000+ institutions2
Monthly new registrations	~65,000	First half of 20262
Monthly clinical consultations	~20 million	As of January 20262
Full-year 2025 revenue	~$150M	+1,803% YoY; ~90% gross margin; ~$124 ARPU2
Most recent funding	$250M Series D	$12B valuation, January 20264

The business model looks more like a search engine than a SaaS seat license. Clinicians use the product free. Revenue comes from pharmaceutical advertising inventory and research partnerships. Healthcare marketing spend is estimated around $30 billion a year, and inventory that reaches physicians directly is scarce and expensive¹. Free clinician access drives density. Density drives advertiser value. OpenEvidence is reported to separate clinical content from ads and to state that ads are not recommendations¹. Whether that line holds under pressure is, in my view, what will decide whether the product stays trusted medical infrastructure or becomes just another attention surface.

Funding, product moves, and EHR integration through early 2026

The funding ladder is easy to lose track of if you only catch one headline. Here is the sequence I can confirm.

Round	Amount	Valuation	Timing	Lead / notable investors
Series A	Size not fully disclosed (cumulative funding over $100M)	$1B	February 19, 2025	Sequoia Capital5
Series B	$210M	$3.5B	July 2025	GV, Kleiner Perkins, Coatue, and others6
Series C	$200M	~$6B	October 2025	Not fully disclosed2
Series D	$250M	$12B	January 2026	Thrive Capital, DST Global4

In about a year the company moved from unicorn status to a $12 billion valuation. The January 21, 2026 Series D brought cumulative funding to roughly $700 million, with annual revenue reported above $100 million⁴. Backers named in coverage include Nvidia, Google Ventures, Sequoia, Blackstone, Bond, Craft Ventures, and Mayo Clinic⁴. If you are still quoting the “$200 million at $6 billion” line as the latest number, that was the Series C and is already one step old.

Usage kept climbing. Monthly consultations moved from about 18 million in December 2025 to roughly 20 million in January 2026. A year earlier the same month was around 3 million, more than a sixfold jump. On March 10, 2026 the platform reportedly processed 1 million consultations in 24 hours⁴². That is not experimental volume.

On the product side, DeepConsult launched with the Series B announcement in July 2025⁶. From one instruction it reads multiple papers in parallel and assembles a doctoral-level research report. The official guide says a single run can use more than 100 times the compute of an ordinary search⁷. The role is shifting from one-shot Q&A to handing off a research task.

Field integration matters more than demo screenshots. At the end of March 2026, Mount Sinai Health System in New York said it would embed OpenEvidence in Epic and roll it out across seven hospitals not only to physicians but also to nurses and pharmacists. Coverage framed that as OpenEvidence’s first enterprise-scale deployment, a step from reference tool to workflow infrastructure⁸. Similar EHR work was reported for Sutter Health in February 2026 and Cedars-Sinai system-wide in May 2026². Once literature search sits inside the chart, the product stops being optional browser tab and starts becoming standard equipment.

Why answers come with citations, and how that actually works

General-purpose systems such as ChatGPT, Claude, or Gemini train on broad internet text. Magazines, novels, forums, and social posts all mix in. OpenEvidence differentiates by tightly narrowing what it treats as source material: peer-reviewed medical literature, society guidelines, and regulated reference sets. Think specialty kitchen, not everything aisle.

Indexed peer-reviewed papers are reported at roughly 35 million¹. The partnerships underneath that corpus are the real moat. On February 19, 2025 OpenEvidence announced a content partnership with NEJM Group covering NEJM, NEJM Evidence, NEJM AI, NEJM Catalyst, and NEJM Journal Watch content and multimedia from 1990 forward⁵. On June 5, 2025 it signed a strategic content agreement with the JAMA Network covering JAMA, JAMA Network Open, and 11 specialty journals⁹. Collaborations also extend to the AMA, NCCN, AAFP, ACEP, and journals such as JAMA Oncology and JAMA Neurology¹. Answers draw on full-text literature, current society guidelines, FDA labeling, CDC infectious-disease guidance, and drug interaction, dosing, and contraindication databases.

Technically the stack is a language model tuned against that corpus plus retrieval-augmented generation. Each question pulls relevant literature and composes an answer with citations a clinician can open on the spot. On August 15, 2025 the company said it became the first AI to score a perfect 100% on the USMLE, after earlier work that first cleared 90%³. Treat exam scores as a signal of literature handling, not as a substitute for clinical judgment.

The design point I care about is not “hallucination never happens.” No current system fully prevents it. OpenEvidence’s choice is to make verification part of normal use by returning sources every time. When something looks off, the physician can check the primary text immediately.

Item	OpenEvidence	General-purpose chatbots
Primary sources	Peer-reviewed medical literature (~35M)	Broad web text
Citations	Shown by default	Optional and uneven
Guideline freshness	Faster via official partners	Bounded by training cutoffs
Primary user	Verified clinicians	General users
Pricing	Free for physicians (core clinical use)	Paid plans common
Regulatory framing	Educational / decision-support tool	Medical use usually discouraged in terms

General LLM medical-task performance has kept rising through 2026. Even so, three things still set OpenEvidence apart in practice: version control of which guideline is being referenced, accountability for citations, and a closed clinician user base.

How clinicians use it, and where it sits next to UpToDate

For U.S. clinicians the onboarding path is straightforward. Create an account at openevidence.com, verify your professional credentials, and start from the web app or the iOS and Android apps. The product is not designed for lay self-diagnosis. It sits in the same professional lane as tools such as UpToDate: material for expert judgment, not a patient-facing diagnostic device.

A typical workflow looks like this. Between patients, a physician enters a chief complaint or a narrow clinical question. The system returns differential candidates, supporting papers, and often suggested next tests, with citations attached. DeepConsult covers longer research tasks when a single outpatient slot is not enough. EHR integrations push the same capability closer to chart review rather than a separate browser habit.

Adjacent products help define the niche. Medical databases such as UpToDate and DynaMed remain strong for curated, expert-edited articles. Paper-summarization tools such as Consensus or Perplexity help with literature scanning. Community-native products such as Doximity GPT and clinical-reasoning tools such as Glass Health occupy nearby seats. Anthropic’s Claude for Healthcare and OpenAI’s healthcare-oriented products are also competing for clinician attention⁴. OpenEvidence’s outline inside that crowd is the combination of clinician verification, mandatory source presentation, and publisher or society partnerships.

Comparison	OpenEvidence	UpToDate
Format	Conversational / agentic	Article database
Sources	Peer-reviewed papers and guidelines	Specialist-authored articles
Updates	Continuous reflection of sources	Article-by-article revision
Price for physicians	Free core clinical use	Paid subscription
Search experience	Ask and get cited answers quickly	Browse structured articles
Strength	Speed and breadth at the point of care	Reliability of expert editing

In practice the two often share the day. OpenEvidence for the three-minute outpatient window. UpToDate when you want to sit with a whole domain. One did not erase the other. That pattern, clinicians using AI while checking sources, is also the pattern companies need when AI touches contracts, policies, and regulated procedures. When we designed ZEROCK, that was the starting point.

International clinicians, including physicians in Japan, have reported successful registration by uploading local medical licenses, and Japanese-language questions with Japanese answers began appearing in 2026 usage reports¹⁰¹¹. Those reports are useful, but they are not a substitute for checking current registration rules on the product itself. Official country-by-country eligibility can change without a blog post.

Safety boundaries U.S. and international teams still need

HIPAA is the first line people ask about. OpenEvidence is reported to have obtained U.S. HIPAA compliance in April 2025¹. That is progress inside the U.S. patient-privacy framework. It is not a blank check to paste chart notes into any AI product. Your hospital or clinic policies, BAAs, and specialty-board expectations still control what may leave the chart. If the clinical question can be asked without identifiers, leave identifiers out. That habit is safer than relying on a compliance badge alone.

A second boundary is regulatory positioning. Literature-referencing educational tools and FDA-cleared diagnostic software are not the same category. OpenEvidence is reasonably treated as an educational or clinical reference aid for professionals, with final judgment remaining the clinician’s responsibility, as its terms repeatedly state. Separate from that, Pathway Labs’ EchoNext, an ECG-based tool for detecting structural heart disease, has been reported to receive FDA clearance with a path toward OpenEvidence integration¹². Do not confuse a literature AI with a cleared diagnostic device. They can sit next to each other in a product story and still live under different rules.

A third boundary shows up for any clinician practicing outside U.S. reimbursement and guideline systems. OpenEvidence preferentially reflects U.S. and partner society evidence. A drug or regimen that is standard in the United States may be unapproved, off-label, or unavailable under another country’s insurance system. The safe habit is the same one good clinicians already use with any U.S.-centric reference: critically appraise the evidence, then translate it against local guidelines and coverage.

Useful domestic guidance for Japanese institutions includes the Ministry of Health, Labour and Welfare’s September 2024 guidelines on using medical digital data for AI research¹³ and the second edition of generative AI use guidelines from HAIP/CIP¹⁴. PMDA SaMD rules still draw a hard line between approved medical-device software and educational tools¹⁵. Those documents are country-specific, but the principle travels: write internal rules that keep AI in an assistive role and make the responsible clinician explicit.

What enterprise knowledge AI can borrow from this design

If you strip OpenEvidence’s clinical packaging, four principles remain.

1. Narrow the reference scope from the open web to a corpus with known quality control.
2. Attach sources so a human can verify the original.
3. Verify users before granting deeper access.
4. Partner with the institutions that own the domain’s source of truth.

Those four transfer to law firms (case law and statutes), manufacturers (technical literature and patents), and financial institutions (regulatory text). The three blockers I hear in enterprise AI projects map almost one-to-one. First, pasting confidential files into general chat tools, even with “no training” contract language, still collides with NDAs and information-security policy. Second, general models do not reliably ground answers in your internal documents, and copy-paste prompting does not scale. Third, answers without sources die at the board or legal review stage. Our monthly-updated AI consulting service WARP exists to walk those issues from framing into a rollout plan.

ZEROCK is TIMEWELL’s product answer for confidential enterprise knowledge. It keeps each customer’s documents in a tenant-isolated scope, attaches source-file links and section references to answers, supports SSO, permissioning, and audit logs, runs on AWS infrastructure in Japan for customers that require domestic hosting, and keeps inputs out of LLM training by contract and architecture. GraphRAG helps with questions that depend on relationships across documents, not only keyword hits. For audit-control design around ISO 42001 and related frameworks, see Enterprise AI Audit Controls.

The binary “ban ChatGPT or allow everything” debate is the wrong frame. Specialized-domain AI is a separate layer. If you are still stuck on that binary, OpenEvidence is worth studying once as a working example of domain corpus, citations, verification, and institutional partnership.

Summary

OpenEvidence is not interesting because it is “AI in medicine.” It is interesting because it productized three hard choices: a closed clinician user base, a literature-first corpus with publisher partnerships, and answers that are designed to be checked. By early 2026 that combination had reached a $12 billion valuation, about $150 million in 2025 revenue, multi-hospital EHR deployments, and agentic research features. U.S. clinicians are the center of gravity. International use is growing, but safety boundaries around patient data, local guidelines, and medical-device regulation still decide whether the tool helps or creates risk.

The transferrable lesson is not “copy a medical AI into your company.” It is rebuild the four principles for your own domain corpus. If that is the problem you are solving, book a consultation and we can map a concrete design for your stack.

Related Articles

• ZEROCK: An Enterprise AI Agent for Confidential Document Knowledge Management
• Enterprise AI Agents Compared: Selection Criteria and Major Domestic Products
• Enterprise AI Glossary: Key Concepts You Should Know

References

Footnotes

1. OpenEvidence Business Breakdown & Founding Story — Contrary Research — accessed 2026-07 — https://research.contrary.com/company/openevidence ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

2. OpenEvidence revenue, valuation & growth metrics — Sacra — updated 2026-05-10 — https://sacra.com/c/openevidence/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

3. OpenEvidence Creates the First AI in History to Score a Perfect 100% on the United States Medical Licensing Examination (USMLE) (official OpenEvidence news release via PR Newswire) — 2025-08-15 — https://www.prnewswire.com/news-releases/openevidence-creates-the-first-ai-in-history-to-score-a-perfect-100-on-the-united-states-medical-licensing-examination-usmle-302531156.html ↩ ↩²

4. OpenEvidence hits $12B valuation with new round led by Thrive, DST — TechCrunch — 2026-01-21 — https://techcrunch.com/2026/01/21/openevidence-hits-12b-valuation-with-new-round-led-by-thrive-dst/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

5. OpenEvidence Achieves $1 Billion Valuation in Sequoia-led Round and Announces Content Partnership with the New England Journal of Medicine (official OpenEvidence news release via PR Newswire) — 2025-02-19 — https://www.prnewswire.com/news-releases/openevidence-achieves-1-billion-valuation-in-sequoia-led-round-and-announces-content-partnership-with-the-new-england-journal-of-medicine-302380960.html ↩ ↩²

6. OpenEvidence announces $210 million round at $3.5 billion valuation (Series B, DeepConsult announcement) — PR Newswire — 2025-07 — https://www.prnewswire.com/news-releases/openevidence-the-fastest-growing-application-for-physicians-in-history-announces-210-million-round-at-3-5-billion-valuation-302505806.html ↩ ↩²

7. DeepConsult user guide — OpenEvidence — accessed 2026-07 — https://www.openevidence.com/user-guide/deep-consult ↩

8. Mount Sinai inks 1st enterprise deal with OpenEvidence for Epic EHR integration — HIT Consultant — 2026-04-01 — https://hitconsultant.net/2026/04/01/mount-sinai-openevidence-ai-epic-integration-nurses-pharmacists/ ↩

9. OpenEvidence and the JAMA Network Sign Strategic Content Agreement (official OpenEvidence news release via PR Newswire) — 2025-06-05 — https://www.prnewswire.com/news-releases/openevidence-and-the-jama-network-sign-strategic-content-agreement-302473690.html ↩

10. What is OpenEvidence? The medical AI used by 40% of U.S. physicians (report on use from Japan) — decades.co.jp — 2026-02 — https://decades.co.jp/openevidence-clinical-guide-202602/ ↩

11. We asked OpenEvidence about long-standing symptoms — GIGAZINE — 2026-06-01 — https://gigazine.net/news/20260601-openevidence-medical-ai-review/ ↩

12. Pathway Labs' EchoNext AI tool for heart disease detection clears FDA — STAT News — 2026-06-23 — https://www.statnews.com/2026/06/23/pathway-labs-echonext-ai-tool-heart-disease-detection/ ↩

13. Guidelines on the use of medical digital data for AI research and development — Ministry of Health, Labour and Welfare — 2024-09 — https://www.mhlw.go.jp/content/001310044.pdf ↩

14. Generative AI Use Guidelines in the Medical and Healthcare Field, Version 2 — HAIP/CIP — 2024-10 — https://haip-cip.org/assets/documents/nr_20241002_02.pdf ↩

15. Medical device programs — PMDA (Pharmaceuticals and Medical Devices Agency) — accessed 2026-07 — https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/0000179749_00004.html ↩