OpenEvidence
The best grounded-answer engine we have scored — and you cannot use it. Free to 915,000 verified U.S. clinicians, blocked in the EU and UK by OpenEvidence's own choice, with no API, no SDK and no MCP for anyone.
PRICING
| Verified U.S. clinician | $0 | unlimited questions, Quick Consult and DeepConsult, web and mobile apps; NPI or proof of student status required |
| EU / UK clinician | Not available | OpenEvidence blocks access in the EU and UK by its own decision, citing the EU AI Act and UK regulatory uncertainty |
| Health system Epic integration | Not published | enterprise deployments announced with Sutter Health, Mount Sinai, Cedars-Sinai, NewYork-Presbyterian; terms undisclosed |
There is no price to negotiate and no subscription to cancel. The thing to price instead is the business model: the tool is free because pharmaceutical advertising pays for it, and the answers a clinician acts on sit on the same screen as that advertising. For any EU buyer the question is moot — the product is geoblocked.
checked 2026-08-13 · vendor pricing page
Element scores
Strengths
Two design decisions carry this product. First, refusal: if the literature does not support an answer, OpenEvidence returns nothing rather than a plausible sentence. In a 2026 otolaryngology study of citation integrity, ChatGPT-4 hallucinated at 23.0% while OpenEvidence produced the most citations per response and the highest mean journal CiteScore. Second, licensed corpora rather than the open web — NEJM, JAMA and all eleven specialty journals, NCCN, Cochrane, ASCO, ACOG, AUA, 300+ journals, FDA and CDC. On top of that sit two transparency layers most vendors never build: 'Why Was This Source Cited', and EvidenceGrade (July 2026), which grades the certainty of the evidence behind each answer A to D on a published GRADE-derived method, and returns U when it cannot grade. Across ~4,500 sampled questions the average grade landed just above a B, with oncology and cardiology strong and geriatrics, genetics and dermatology weak — the company published the weak result as well as the strong one. Adoption backs the design: 915,000 verified U.S. clinicians as of July 2026, more than 40% of U.S. physicians logging in daily, ~18M consultations in December 2025, $250M Series D at a $12B valuation in January 2026, roughly $300M annualised revenue by mid-2026.
Honest dings
For anyone reading this from Belgium, France or the UK, the audit ends at one line on OpenEvidence's own site: it is not available in the European Union or the United Kingdom, because of the EU AI Act and UK regulatory uncertainty. That is not a verification friction you can work around with a VPN — it is a deliberate market withdrawal, and it is the single most important fact about this tool for a European buyer. Second: this is a closed box. No public API, no SDK, no official MCP, no A2A. The only MCP route is an unofficial community server that drives your own browser session. Building anything on top of OpenEvidence means an enterprise Epic negotiation. Third, the accuracy picture is not uniform. A December 2025 medRxiv pilot on MedXpertQA-derived subspecialty scenarios scored DeepConsult at 41% and Quick Consult at 34% — the authors noted all models scored poorly on that dataset, but the gap between board-exam performance and hard subspecialty reasoning is real. The npj Digital Medicine systematic review (12/08/2026, eleven studies) reached the same shape: strongest on guideline-based questions, variable on complex scenarios, and the platform mostly reinforced clinical decisions rather than changing them. Fourth, it is ad-supported. Pharmaceutical advertising sits next to clinical answers. Fifth, it does not do differential diagnosis generation or drug dosing, so it is a lookup layer, not a bedside decision-support suite.
Sources (16) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- openevidence.com/user-guide/ask-overview — Official product documentation: 100M+ AI-powered clinical consultations completed to date; grounding in NEJM, JAMA, NCCN, Cochrane; stated use cases (quick consults, differential, treatment options, drug info, guideline summaries, prior authorizations, patient education) (accessed 2026-08-13)
- openevidence.com/user-guide/clinical-curbside — PRIMARY EVIDENCE for the EU/UK block, stated in OpenEvidence's own words: 'Due to mounting regulatory uncertainty regarding the treatment of AI systems in the European Union and the United Kingdom, including, among other rules, the EU Artificial Intelligence Act, OpenEvidence is not available in the European Union or the United Kingdom.' (accessed 2026-08-13)
- openevidence.com/register — Official registration page: free for verified U.S. healthcare providers, verified-HCP-only access, SOC 2 Type II (accessed 2026-08-13)
- openevidence.com/blog/introducing-evidencegrade-gra… — Official methodology post (09/07/2026): full EvidenceGrade design — two-phase scoring, A/B/C/D grades with +/- modifiers and U for ungradeable, GRADE-derived ceiling-from-study-design logic, deference to expert synthesis, plus the ~4,500-question sample result (average just above B; oncology/pharmacy/endocrinology/cardiology strong, geriatrics/genetics/dermatology weak) and the explicit admission that grades will be wrong at the edges (accessed 2026-08-13)
- nature.com/articles/s41746-026-03077-4 — PRIMARY INDEPENDENT EVIDENCE. Artsi, Sorin, Glicksberg, Freeman, Korfiatis, Bratt, Nadkarni & Klang, 'OpenEvidence clinical question-answering platform: systematic review of early evaluations', npj Digital Medicine, 12/08/2026, doi:10.1038/s41746-026-03077-4 — eleven included studies; lower fabricated-citation rates than general-purpose LLMs; strongest in guideline-based settings, variable in complex scenarios; platform often reinforced rather than altered clinical decisions; evidence base limited by small samples and methodological heterogeneity (accessed 2026-08-13)
- medrxiv.org/content/10.64898/2025.11.29.25341091v1.… — Independent preprint (Jagarapu et al., medRxiv, 29/11/2025): accuracy and repeatability of OpenEvidence on complex medical subspecialty scenarios from the MedXpertQA dataset; 41% DeepConsult / 34% Quick Consult; authors note suitability for subspecialty questions is unclear and that references were not exhaustively checked for hallucination (accessed 2026-08-13)
- link.springer.com/article/10.1007/s00405-026-10253-5 — Independent peer-reviewed: diagnostic accuracy and citation integrity of four LLMs on otolaryngology vignettes — OpenEvidence and Perplexity generated the most citations per response, ChatGPT-4 the fewest with the highest hallucination rate (23.0%), OpenEvidence highest mean journal CiteScore (accessed 2026-08-13)
- fiercehealthcare.com/ai-and-machine-learning/openev… — Independent (Fierce Healthcare, 09/07/2026) reporting company figures: 915,000 licence-verified U.S. clinicians, 690,000+ licence-verified U.S. physicians as of July 2026; EvidenceGrade launch and grading bands (accessed 2026-08-13)
- fiercehealthcare.com/ai-and-machine-learning/openev… — Independent (Fierce Healthcare, 21/01/2026): $250M Series D at $12B valuation co-led by Thrive and DST, ~$700M raised in a year; ~18M consultations in December 2025 vs ~3M/month a year earlier; 40%+ of U.S. physicians daily across 10,000+ hospitals; free and ad-supported; topped $100M annual revenue; multi-AI agentic architecture and subspecialist-agent roadmap (accessed 2026-08-13)
- openevidence.com/announcements/openevidence-partner… — Official (20/05/2026): Cedars-Sinai partnership integrating Epic patient context directly into OpenEvidence; agentic system gathers patient data, evaluates literature and synthesises; EHR data not stored after the clinical session (accessed 2026-08-13)
- openevidence.com/announcements/openevidence-the-fas… — Official: DeepConsult wide release — autonomous PhD-level research agent, 100x the compute and cost of a standard search, free to all verified U.S. clinicians; content partnerships with AMA, NEJM, JAMA and all eleven JAMA specialty journals (accessed 2026-08-13)
- nbcnews.com/tech/tech-news/openevidence-ai-doctor-m… — Independent (NBC News, 13/05/2026): access requires a U.S. government-issued NPI; unlimited questions once registered; free (accessed 2026-08-13)
- clinicalaireport.com/reviews/open-evidence — Independent clinician-authored review (08/05/2026): 81/100 from 16 physician reviews; no differential diagnosis generation and no drug dosing tools; ad-supported model flagged as a procurement consideration; EHR integration described as early-stage (accessed 2026-08-13)
- github.com/bakhtiersizhaev/openevidence-mcp — Evidence that there is no official developer surface: the only OpenEvidence MCP server is an unofficial open-source project that uses a Playwright browser session because 'OpenEvidence lacks a public API token flow' (accessed 2026-08-13)
- research.contrary.com/company/openevidence — Independent research profile: 35M+ peer-reviewed publications indexed; specialised smaller medical models rather than a general-purpose LLM; no-answer-when-evidence-is-inconclusive policy; free with NPI verification (accessed 2026-08-13)
- iatrox.com/blog/openevidence-outside-us-access-veri… — Independent corroboration of the international access problem from a UK clinician's perspective (published 27/12/2025): NPI-based verification has no non-US pathway; UK/EU clinicians locked out of full features (accessed 2026-08-13)