Anthropic Bought What OpenAI Couldn't Win — Direct Distribution to 757,000 U.S. Doctors

On September 22, 2026, Anthropic and OpenEvidence announced a global clinical-decision-support partnership that Reuters carried under an \"EXCLUSIVE\" banner. Under the agreement, Anthropic supplies the back-end AI for a regionally adapted, free version of OpenEvidence that will roll out across roughly 100 low- and middle-income countries — a list that includes Uganda, Angola, Sudan, Haiti, and Mongolia. OpenEvidence does the clinical tailoring; the model plumbing is Anthropic's.

Financial terms were not disclosed, but the strategic value is hard to overstate. OpenEvidence already hosts 757,000+ verified U.S. clinicians and is used daily by more than 40% of American physicians — in August 2026 alone, U.S. clinicians logged 42 million clinical consultations on the platform. That footprint puts a free, Anthropic-powered clinical AI in front of more practicing doctors than any other frontier-model vendor in the world, ahead of the IPO prospectus that Anthropic quietly filed with the SEC on June 1, 2026.

Read the headline that framed this deal and you would think 757,000 doctors are now pinging Claude every time they have a clinical question. They are not. On September 3, 2026 — nineteen days before the Anthropic announcement — OpenEvidence shipped its own model family: Osler (the new default), Sackett, Snow, and a research-preview model called Darwin that scored a perfect 660/660 on a physician-reviewed MedQA test set and 82.7% on HealthBench Professional, 12.1 points above the next-best model. The physician's first question goes to OpenEvidence's stack, not to Claude.

Where Claude does show up is in the grading and back-end layer. OpenEvidence's own benchmark methodology explicitly uses Claude Opus 4.8 to grade HealthBench Professional responses \"mirroring the grading configuration Anthropic published with the Claude Fable 5 release,\" per the company's September 3 post. The LMIC partnership extends that relationship: Anthropic provides the underlying infrastructure for the globally distributed, regionally adapted version, while OpenEvidence owns the medical-specialist surface and the physician relationship.

Anthropic exited 2025 at roughly $9B in run-rate revenue, hit $47B by May 2026, and reached $65B by the end of July 2026 — a 7x jump in seven months that, on its own, justifies the $965B valuation it printed in its $65B Series H round led by Altimeter, Dragoneer, Greenoaks, and Sequoia Capital. But run-rate alone does not sell an IPO narrative. S-1 buyers want visible, durable end-markets where the AI is locked in at the workflow level, not just rented at the API.

Healthcare is the obvious vertical. Anthropic's pre-IPO healthcare stack now includes Claude for Healthcare (launched January 11, 2026, with HIPAA-ready connectors to CMS Coverage, ICD-10, NPI Registry, and PubMed), Claude for Life Sciences, Claude Science ($400/month researcher workbench), a $400M all-stock acquisition of biotech AI startup Coefficient Bio in April 2026, and a wet lab in the San Francisco Bay Area that Anthropic quietly stood up in mid-September. Add the Optum partnership announced in July 2026 — a $3B AI capitalization program inside UnitedHealth Group's claims-processing and revenue-cycle business — and the OpenEvidence deal is the missing piece that closes the loop from bench to bedside.

Three weeks before the OpenEvidence deal, on September 1, 2026, OpenAI announced the opposite distribution strategy: a direct integration with Epic that gives ChatGPT read-only access to 325 million patient records, with UCSF Health as the pilot site. The framing is patient-data moats. The friction is procurement: Epic integrations move at CIO speed, and even after integration, ChatGPT still depends on Epic's permission to keep working with that data.

Anthropic skipped the gatekeeper entirely. OpenEvidence sells no EHR integration — its whole growth engine is bypassing hospital procurement by giving verified physicians a free, mobile-first app. In2025 alone, 100 million Americans were treated by a doctor who used OpenEvidence to help guide care, per the company's disclosures. That is the distribution Microsoft-backed OpenAI cannot replicate without its own consumer-grade physician app — and the only competitor in that lane is Doximity, whose Q1 FY2027 revenue of $156.6M (up 7.3% year-on-year, reported August 6, 2026) shows the audience is reachable but still smaller than OpenEvidence's monthly-consultation count.

The clinical-AI market is consolidating into three distinct distribution strategies, and each maps to a different public-market story. The EHR-integrated incumbent: Wolters Kluwer's UpToDate Expert AI is now live in roughly 2,500 U.S. hospitals and health systems — more than 90% of Wolters Kluwer's UpToDate footprint — and added CME-credit delivery in2026 to deepen lock-in. Its moat is enterprise contracts and 20,771 employees of compliance-grade content; its weakness is slow consumer adoption outside the hospital firewall.

The physician-social incumbent: Doximity generates $644.9M of FY2026 revenue (ended March 31, 2026, up 13% year-on-year, reported May 13, 2026) and is now a credible AI suite, with management saying nearly half of its provider base used clinical AI in Q4 FY2026. The stock trades at 7.7x trailing sales, a discount that reflects growth deceleration as OpenEvidence steals physician mind-share. Alphabet's AMIE is the research-system wild card: a Nature Medicine study in June 2026 showed AMIE matching primary-care physicians on multi-visit disease management and beating them on text-based simulated cases, with a real-patient primary-care pilot already deployed. AMIE is research-only today; Alphabet has yet to commercialize it at OpenEvidence's scale.

Short-term (days to quarters): the next milestone is Anthropic's IPO prospectus becoming public — currently targeted for October 2026 on Nasdaq, though one Schwab Network report cites a push to November. The S-1 will quantify healthcare's share of the $65B run-rate and confirm whether the OpenEvidence LMIC deployment shows up as a revenue line (free, so likely a cost) or as a goodwill narrative. OpenEvidence is reportedly exploring a $200M round at a $20B valuation (up from $12B in January 2026) — a closing would reset the comp set for physician-AI private valuations.

Long-term (one to three years): the structural question is whether Anthropic can hold the rails under OpenEvidence's expanding model family. Darwin's perfect MedQA score is a competitive threat to anyone who wants to charge for clinical AI — including Anthropic's own Claude for Healthcare product. If OpenEvidence goes public or signs a competing hyperscaler, Anthropic loses both the distribution and the evaluation layer. The most credible risk is also the most boring: OpenAI ships a free consumer-grade clinical AI in 2027, Microsoft bundles it with Epic on Azure, and physician default-habit re-anchors before Anthropic's IPO lock-up expires.