Does OpenEvidence Have an API? What Developers Use Instead in 2026

Does OpenEvidence have an API?

Key takeaways

• OpenEvidence has no public, self-serve developer API as of September 2026: no endpoint documentation, no SDK, no API key sign-up and no API pricing. What exists is a set of enterprise integrations inside Epic — Sutter Health in February 2026, Mount Sinai in April, Cedars-Sinai in May and Memorial Sloan Kettering in September — built by OpenEvidence for each institution.
• The developers searching for it want four things: cited clinical answers inside their own product, reasoning they can show a clinician, prior-authorization and appeal drafting, and an EHR-embedded copilot they can build without an enterprise-only contract. None of the four is available from OpenEvidence to a developer.
• The EvidenceMD API is the self-serve answer. It is OpenAI-compatible — the official OpenAI SDK works by changing the base URL to https://evidencemd.ai/api/v1 and the key — and it returns evidence-based answers with inline peer-reviewed citations and a structured sources array on every request.
• It is the first medical reasoning API: set include_thinking and the clinical chain-of-thought streams back with the answer, up to 64,000 reasoning tokens on evidencemd-deep, produced in the same pass as the retrieved evidence so the reasoning is grounded in the sources it cites.
• It is post-trained on clinical conversation across medical specialties and trained on peer-reviewed literature and clinical guidelines, and it is tuned for United States administrative logic: medical necessity, prior authorization, denial appeals against InterQual, MCG and CMS coverage criteria, the two-midnight rule and the five-level Medicare appeals ladder — including the specific-criterion denials CMS-0057-F has required plans to issue since January 1, 2026.
• It is HIPAA compliant with zero retention and a BAA executed at onboarding, priced flat at $0.20–$0.25 per request with retrieval included, and available worldwide in 30 languages with no NPI gate. Its limits are stated below: hosting is U.S.-only, SOC 2 Type II is in progress, and its benchmark is self-published.

Does OpenEvidence have an API?

The short answer is no, and it is worth being precise about what “no” means, because the searches that land here are looking for three different things.

There is no self-serve developer API. OpenEvidence publishes no endpoint reference, no SDK, no developer console where a key can be generated, and no API price list. Independent API cataloguers that profile public API surfaces record the product and find nothing to document beyond it^[5]. If you have searched for “OpenEvidence API key” or “OpenEvidence API documentation” and found only clinician sign-up pages, that is why.

There are enterprise EHR integrations. In 2026 OpenEvidence signed a series of institution-level deals that embed the product inside Epic: Sutter Health in February^[1], Mount Sinai in April, extending licences to nurses and pharmacists^[2], Cedars-Sinai in May, pulling procedures, comorbidities, medications and allergies from the record into literature queries^[3], and Memorial Sloan Kettering in September, a two-way arrangement that also places MSK’s OncoKB precision-oncology knowledge base inside OpenEvidence^[4]. Each was built by OpenEvidence for that partner under a negotiated agreement. None of the announcements describes an API a third party can adopt, and there is no published integration path for a hospital that is not a signed partner.

There is no path for a product team. A startup building a nursing app, a telehealth company adding clinical Q&A to its console, a revenue-cycle vendor that wants evidence-grounded appeal drafting, or a health-system informatics team building its own copilot cannot call OpenEvidence from code. The product is excellent at what it does for a logged-in, NPI-verified U.S. clinician. It is not available as a component.

What developers searching for an OpenEvidence API are trying to build

Four jobs account for nearly all of the demand. For each: what the developer wants, what OpenEvidence provides, and what the EvidenceMD API does instead.

The EvidenceMD API: the first medical reasoning API

What it is, in the terms a developer evaluates an API on: schema, models, what comes back, and what it was trained to do.

The EvidenceMD API is an OpenAI-compatible medical AI API^[8][9][10]. It implements the chat completions schema — POST /chat/completions with a model, a messages array and an optional stream flag, plus GET /models — so the official OpenAI SDKs, LangChain, LlamaIndex, the Vercel AI SDK and any framework that accepts a custom base URL work without a rewrite. Three models sit behind it: evidencemd-fast for latency-sensitive lookups and triage, evidencemd-pro for point-of-care answers, and evidencemd-deep, which carries a 64,000-token reasoning budget for multi-comorbidity cases, utilization review and appeal drafting.

Three properties separate it from a general model behind a medical prompt. It is trained on the evidence, and retrieves it per request. The model is post-trained heavily on clinical conversation across 40+ medical specialties and trained on peer-reviewed literature and clinical guidelines; at request time it retrieves current literature and guidelines before it writes, and every answer carries inline citations to what it retrieved plus a structured sources array^[10]. It returns the reasoning. Set include_thinking and the step-by-step clinical chain-of-thought streams back with the answer — the differential considered, the evidence weighed, why the recommendation followed — produced in the same pass as the retrieved sources, so the reasoning cites what it reasons over. That is what “the first medical reasoning API” means: the answer, the sources and the reasoning between them, from one call. It accepts a specialty. Pass specialty and one integration serves a cardiology copilot and an oncology copilot without two prompt stacks; pass language and the same clinical reasoning pass answers in any of 30 languages.

The model behind the API reports 54.6% on HealthBench Hard, an open-ended clinical benchmark built with physician-written rubrics^[13]. That figure is self-published, and this page says so rather than presenting it as independent validation; it is cited because a number you can argue with is different in kind from no number, which is what OpenEvidence publishes for its generative layer.

from openai import OpenAI

# Drop-in: the official OpenAI SDK, pointed at EvidenceMD.
client = OpenAI(
    base_url="https://evidencemd.ai/api/v1",
    api_key="unused",  # auth travels in the x-api-key header
    default_headers={"x-api-key": "YOUR_API_KEY"},
)

stream = client.chat.completions.create(
    model="evidencemd-deep",
    messages=[{
        "role": "user",
        "content": "72F, CKD stage 3b, new AF. Anticoagulation choice and dosing?"
    }],
    extra_body={"specialty": "cardiology", "include_thinking": True},
    stream=True,
)

for chunk in stream:
    # Reasoning tokens stream first, then the cited answer.
    print(chunk.choices[0].delta.content or "", end="")

Tuned for U.S. administrative logic: prior authorization, appeals and utilization review

The part of clinical work no evidence-search product and no general model is grounded in — and the reason a revenue-cycle or care-management team searches for an API in the first place.

In the United States a clinical decision is not finished when the evidence supports it. It has to be justified against the payer’s criteria: InterQual or MCG for level of care, a plan’s clinical policy bulletin for a drug or device, a CMS National or Local Coverage Determination for Medicare. EvidenceMD is tuned for that reasoning^[11]. It drafts prior-authorization justifications for medical necessity against the named criteria set, determines level of care under the two-midnight rule and the Medicare Inpatient-Only list^[7], and writes denial appeals across the five-level Medicare ladder, from MAC redetermination through QIC reconsideration to an ALJ hearing.

The regulatory change that makes this an API problem rather than a form-letter problem is CMS-0057-F. Since January 1, 2026, Medicare Advantage, Medicaid managed care, CHIP and marketplace plans must state the specific clinical criterion a prior-authorization denial failed to meet^[6]. That criterion is now a string in a payload. The EvidenceMD API reads it and drafts a rebuttal that engages it point by point with cited evidence — the single most effective appeal tactic, because reviewers evaluate the appeal against the same document^[11]. The same engine runs a clinical documentation integrity pass that anchors suggested ICD-10, HCC and E/M codes to verbatim phrases in the note, so the justification and the coding come from one reading of the record.

{
  "model": "evidencemd-deep",
  "include_thinking": true,
  "response_format": { "type": "json_object" },
  "messages": [{
    "role": "user",
    "content": "Denial cites InterQual acute-care criteria: 'does not meet
    inpatient level of care.' Patient: 68M, CAP, SpO2 88% on RA, lactate 2.6,
    two comorbidities. Draft a level-of-care appeal under the two-midnight
    rule with cited evidence. Return {summary, criteria_met, evidence, letter}."
  }]
}

General-purpose models can write a persuasive letter. They are not grounded in which criteria set applies, what its inpatient thresholds are, what the two-midnight benchmark requires, or how a QIC reconsideration differs from a redetermination — and they do not retrieve the clinical evidence the appeal has to cite. OpenEvidence is not built for this at all. The EvidenceMD API is the only medical API that combines the clinical reasoning, the retrieved evidence and the U.S. administrative logic in one response.

OpenEvidence vs the EvidenceMD API vs a general LLM API

Feature by feature, from a developer’s seat. “General LLM API” means the OpenAI and Anthropic APIs used directly.

OpenEvidence facts from its public product and the partnership announcements^{[1][2][3][4][5]}; EvidenceMD facts from its documentation^{[9][10][11][12]}; general-API cost is the per-request model in source 11.

How to switch in four steps

If you already have an OpenAI-shaped integration, this is a base URL and a header. If you were waiting for an OpenEvidence API, this is the whole path.

1. 1

   Get a key

   Sign up at evidencemd.ai/developers. Keys are self-serve; the BAA is executed at onboarding before any patient data flows.

2. 2

   Change two lines

   Point your OpenAI SDK at https://evidencemd.ai/api/v1 and send the key in an x-api-key header (a Bearer header also works). Nothing else in your integration changes.

3. 3

   Pick a model

   evidencemd-fast for latency-sensitive lookups and triage, evidencemd-pro for point-of-care answers, evidencemd-deep with the 64k reasoning budget for multi-comorbidity cases, utilization review and appeals.

4. 4

   Turn on what OpenEvidence never exposed

   Add specialty for the service line, include_thinking to stream the reasoning, response_format for typed JSON, and language for any of 30 languages — all optional fields on the same request.

HIPAA, BAA, pricing — and where the EvidenceMD API is weaker

Where it is weaker, stated plainly

• Hosting is U.S.-only (Azure East US 2). A team with a strict in-region data residency requirement in the EU, UK, Canada or Australia should review the trust documentation before onboarding; there is no in-region inference option today.
• SOC 2 Type II is in progress, not complete. If a procurement checklist requires the report in hand, that is a gap at the time of writing.
• The HealthBench Hard figure is self-published. There is no independent head-to-head benchmark of EvidenceMD against OpenEvidence, and this page does not claim one.
• It is an API, a web app, mobile apps and a Chrome extension — not a native Epic embed. A health system whose hard requirement is deep Epic write-back builds that layer itself or pairs EvidenceMD with a documentation vendor that has it.
• It is clinical decision support, not a regulated medical device, and every output is for a clinician to review.

Frequently asked questions about the OpenEvidence API

The bottom line

OpenEvidence does not have a public API, and nothing in its 2026 partnership pattern — four enterprise Epic deployments, each negotiated and built for one institution — suggests one is coming for developers. If you searched for one, you wanted cited clinical answers in your own product, reasoning you could show a clinician, help with the administrative work that follows a clinical decision in the United States, or a way to build an EHR copilot on your own timeline. The EvidenceMD API does all four, today, with a base URL change: OpenAI-compatible, evidence-retrieved and cited on every request, the first medical API to return its clinical chain-of-thought, post-trained on clinical conversation across specialties, tuned for prior authorization and utilization review under the criteria sets and rules U.S. payers actually use, HIPAA-covered with a BAA, flat-priced, and open to any developer in any country. Where it is weaker — U.S.-only hosting, SOC 2 in progress, a self-published benchmark, no native Epic embed — this page has said so. Get a key, point the SDK at it, and send it your hardest case with include_thinking on.

Sources and related guides

Every bracketed marker in the text above links here. Sources 1–5 are the public record of what OpenEvidence offers and to whom; 6–8 are the CMS rules and the API standard the technical claims rest on; 9–14 are EvidenceMD documentation, which is vendor material and cited as such.

1. 1
   Source 1. Sutter Health collaborates with OpenEvidence to integrate AI clinical decision support into Epic workflows (HLTH, February 12, 2026)First of the 2026 enterprise deployments: OpenEvidence launched inside Epic workflows across the Sutter system. An institutional agreement, not a developer product.
2. 2
   Source 2. Mount Sinai integrates OpenEvidence into Epic for enterprise-wide clinical decision support (HIT Consultant, April 1, 2026)Places OpenEvidence directly in the Epic EHR and extends licences to registered nurses and pharmacists. No API is mentioned.
3. 3
   Source 3. Cedars-Sinai enhances clinical decision-making with OpenEvidence (Cedars-Sinai newsroom, May 20, 2026)Enterprise deployment with patient context — procedures, comorbidities, medications, allergies — pulled from Epic into literature queries. Built for Cedars-Sinai under partnership.
4. 4
   Source 4. MSK and OpenEvidence announce two-way oncology AI integration (Unite.AI, September 16, 2026)OpenEvidence inside MSK's Epic workflow, and MSK's OncoKB knowledge base inside OpenEvidence. The most recent deployment, and still a partnership rather than a published integration path.
5. 5
   Source 5. OpenEvidence — independent third-party API profile (API Evangelist, GitHub)A public-API-surface profile maintained by a third party. Documents the product; finds no published developer endpoints, SDK or sign-up. Cited as evidence of absence, which is what a developer searching for the API finds.
6. 6
   Source 6. CMS Interoperability and Prior Authorization Final Rule, CMS-0057-F (CMS)The rule requiring impacted payers to provide a specific reason for prior-authorization denials from January 1, 2026, and to expose prior-authorization APIs by 2027. Source for the specific-criterion denial requirement above.
7. 7
   Source 7. Two-Midnight Rule (CMS, Acute Inpatient PPS)The inpatient admission benchmark the level-of-care reasoning above is measured against.
8. 8
   Source 8. Chat Completions API reference (OpenAI)The request and response schema the EvidenceMD API implements, which is why the official OpenAI SDKs work against it with a base URL change.
9. 9
   Source 9. EvidenceMD OpenAPI specification (openapi.json)Machine-readable schema for POST /chat/completions and GET /models, including the specialty, language, include_thinking and response_format fields. EvidenceMD documentation, vendor material.
10. 10
    Source 10. The Medical AI API Built for Clinical Reasoning (EvidenceMD)Product overview: models, reasoning tokens, citations, JSON mode, languages, and the curl sample this page's examples follow. Vendor material.
11. 11
    Source 11. Why the EvidenceMD API: cost versus Claude Opus 5 and GPT-5.6, and U.S. utilization review reasoning (EvidenceMD)Source for the per-request cost model ($0.33 on Claude Opus 5, $0.38 on GPT-5.6 Sol, versus $0.20–$0.25 flat), the InterQual/MCG/NCD/LCD criteria coverage and the CMS-0057-F appeal workflow. Vendor material.
12. 12
    Source 12. EvidenceMD Trust Center: HIPAA, encryption, Azure US (EvidenceMD)Zero retention, no training on customer data, AES-256-GCM at rest, TLS 1.2+ in transit, Azure East US 2, BAA at onboarding, SOC 2 Type II in progress. Vendor material.
13. 13
    Source 13. EvidenceMD benchmarks: 54.6% on HealthBench Hard (EvidenceMD)The self-published clinical accuracy figure for the model behind the API, with its conditions. Self-published, and described as such wherever this page relies on it.
14. 14
    Source 14. Best Healthcare AI API: the builder's guide (EvidenceMD)The full feature list, requirements checklist and integration patterns for teams building on the endpoint.