This is not a leaderboard. UpToDate Expert AI, OpenEvidence Ask, Dyna AI, Glass, iatroX and Meddies overlap at the edges, but they do not all solve the same job. Some answer questions from a reference corpus. Glass also drafts encounter notes. OpenEvidence Visits is a separate encounter-documentation surface; the OpenEvidence row below covers Ask. Meddies is being built around patient data and hospital workflow.
The comparison below uses public vendor pages reviewed in July 2026 and the current Meddies codebase. A vendor page can establish what a product says it contains or supports. It cannot establish that one product is safer, more accurate or more effective than another without a head-to-head evaluation.
What each public product page establishes
| Tool | Publicly described content boundary | Source behavior described publicly | Workflow and access information |
|---|---|---|---|
| UpToDate Expert AI | UpToDate says Expert AI is built on its curated clinical corpus and evidence-based recommendations from more than 7,600 clinicians [1] | Responses include inline links to supporting UpToDate topics [1] | Included in select personal and enterprise subscriptions; search supports 16 languages, while clinical content remains in English [2] |
| OpenEvidence Ask | The OpenEvidence site lists full-text content partnerships with NEJM, JAMA, Cochrane and NCCN, plus multiple medical societies [3] | The public page reviewed does not specify the citation format for every Ask response | Ask is the clinical-question surface compared in this row; OpenEvidence Visits covers a separate encounter-documentation workflow. The public page reviewed does not state a complete price or geographic-access policy [3] |
| Dyna AI | EBSCO says Dyna AI retrieves only from DynaMedex, DynaMed or Dynamic Health [4] | The product identifies the originating content set and provides direct source links [4] | DynaMed lists EHR integration; the Dyna AI page offers institutional contact and a US-resident free trial [4] [5] |
| Glass | Glass describes an evidence base of peer-reviewed literature, clinical guidelines and US FDA drug information [6] | Its public page describes responses grounded in medical evidence with citations [6] | Combines ambient scribing and clinical decision support; its pricing page lists EHR integration for Athena, Epic, eCW and Elation on the Max tier [7] |
| iatroX | The vendor says its clinical question tool uses NICE, CKS, SIGN and SmPC sources [8] | The same page describes answers with citations; its comparison article calls the product UK-guideline-first [8] [9] | The clinical question tool is free; paid subscriptions cover additional exam-question banks [8] |
| Meddies | The current implementation can retrieve FHIR patient data, build a structured patient summary and query a drug-interaction source; a public clinical corpus has not been specified | Clinical instructions require citations for recommendations and provenance for patient-summary fields. This is a design contract, not a measured guarantee | Current code contains a FHIR-backed patient view and patient-summary action. The intended deployment model is inside hospital workflow; pricing is not public [10] |
The rows are less comparable than they look
“Has citations” can describe several different things. UpToDate links an AI response to an UpToDate topic. Dyna AI links to its curated content sets. Glass describes citations to medical literature. iatroX says its answers cite UK sources. OpenEvidence publishes a detailed list of content partnerships, but the public page reviewed does not document the citation contract for every OpenEvidence Ask response.
Those are useful facts. They do not show whether a cited passage supports the sentence beside it, whether the source is current for the patient’s setting, or how often the model abstains when evidence is thin. A citation feature still needs claim-level verification.
Workflow claims also need careful wording. Wolters Kluwer’s 2025 APAC announcement says UpToDate Enterprise Edition was in use at more than 200 hospitals across Australia, New Zealand, Singapore, Taiwan and Hong Kong. That announcement does not prove the product is unavailable elsewhere. Glass and DynaMed publish integration options, but an integration label does not tell us which patient fields are read, written or available at decision time.
The same caution applies to regulation. iatroX states that its UK-guideline Q&A tool is MHRA-registered and UKCA-marked as a Class I medical device. Its Q-banks and other learning tools are not medical devices. That vendor-reported status does not show that the Q&A tool outperforms the other products in this table.
Meddies has a narrower current claim
The current Meddies code supports two relevant paths. Its drug-interaction tool accepts a medication list and returns resolved drug names, pairwise interactions, severity labels and descriptions from Medscape. Its patient-summary workflow retrieves FHIR data and instructs the model to record provenance, mark missing data and check medications, allergies and recent clinical trends before presenting a summary.
The code does not establish that every hospital EMR is integrated, every answer has a clickable source, every interaction is clinically relevant, or either workflow improves patient outcomes. Those remain product and evaluation work. So the Meddies column describes current components and intended placement, not a clinical-performance result.
This distinction also changes how to read the other columns. Vendor documentation describes capabilities under each vendor’s conditions. It is not a shared benchmark, and no row here supports a claim that one tool is the safest choice for a Vietnamese hospital.
What a Vietnamese hospital still has to ask
The first question is the job. Does the clinical team need a reference answer, a literature search, an encounter note, a drug-interaction check or a summary built from the patient record? Buying one category while expecting another creates a mismatch before accuracy is even measured.
Next comes the evidence boundary. Which guidelines, drug sources and local policies can the system use? Can a clinician open the source behind a specific claim? What happens when the answer depends on information missing from the record? Which data leave the hospital, and what is written back to the EMR?
Finally, test the tool in the intended workflow. Public product pages cannot substitute for local validation with the hospital’s data, users and safety process.
Circular 13/2025/TT-BYT set deadlines for implementing electronic medical records: 30 September 2025 for licensed hospitals and 31 December 2026 for other facilities providing inpatient, day and outpatient treatment. It did not mandate a CDSS, approve any tool in this comparison or prove that an AI layer is ready for clinical use.
The schedule creates an electronic-record base. Choosing what should reason over that base remains a separate decision, and the honest comparison begins by keeping those two decisions apart.