How to check whether a medical AI's references support its answer
A practical method for checking that a paper exists, matches the claim and supports the conclusion being made.
Read the guideA fresh MontMedical Pro Model review, alongside Fable 5.2 Max, DeepSeek Flash Max, Gemini Carbon High and OpenEvidence. Compare the ratings and the actual words.
17 November 2026 · Selected development cases · Internal diagnostic-reasoning review
Ark Medical has been evaluating the MontMedical Pro beta. This review compares its fresh diagnostic answers across all 52 complete cases with retained OpenEvidence answers and direct Fable 5.2 Max, DeepSeek Flash Max and Gemini Carbon High responses.
| Tested profile | Mean score |
|---|---|
| MontMedical Pro ModelMontMedical Pro beta | 9.48 /10 |
| Fable 5.2 MaxAnthropic Fable 5.2 · Max (adaptive thinking) reasoning | 8.94 /10 |
| DeepSeek V4.1 Flash 1017DeepSeek-V4.1-Flash-1017 · Max reasoning | 8.73 /10 |
| Gemini 4 Carbon HighGoogle Gemini 4 Carbon · High reasoning | 8.55 /10 |
| OpenEvidencePreviously supplied OpenEvidence answers | 8.18 /10 |
Fable returned 49 diagnostic answers from 52 submitted cases. Its mean across all 49 answers is 8.96/10; its shared-case mean above is 8.94/10 across the same 48 cases as the other profiles. Cases 32, 43 and 48 are marked Provider error, with no score. Case 35 has no supplied OpenEvidence answer. All four are excluded from every shared mean. Existing case scores are unchanged; the comparison denominator is now 48.
Selected development cases. Internal, AI-assisted beta evaluation of diagnostic reasoning, not measured clinical accuracy. The systems used different prompts, reasoning settings and evidence access. These selected-case scores do not establish broad clinical superiority.
Exact models and access routes → All case ratings → Method, disclosure and limits →
A practical method for checking that a paper exists, matches the claim and supports the conclusion being made.
Read the guideBuild a concise, chronological case summary that preserves exact findings, treatment timing and the question you need answered.
Read the guideUnderstand reference ranges, pre-test probability and why the specimen, method and conditions matter when interpreting a reassuring result.
Read the guideSeparate biological activity, the cells carrying a variant, disease classification and evidence of benefit from a targeted treatment.
Read the guideAn organised, respectful approach to preparing a timeline, preserving your parent’s priorities and leaving with a clear follow-up plan.
Read the guideA transparent comparison looks beyond confident language: inspect factual fidelity, references, reasoning, next steps and performance on unseen cases.
Read the guideInvented cases. Real questions about how an observation becomes a conclusion.
A synthetic immune-function case showing why the stimulus, cell type and biological compartment change what a test establishes.
Follow the reasoning →A synthetic JAK3 case separating a published disease association from cellular assignment in a mixed sample and evidence for a targeted treatment.
Follow the reasoning →An explanation is more useful when you can see the finding, the source and the inference connecting them. Our guides keep those steps separate and explain why the distinction matters.
Ark Medical articles are AI-assisted and source-linked; independent clinical peer review is not claimed. Read our editorial approach →
Ark Medical has been examining the MontMedical Pro beta: how it recognises a diagnosis, interprets tests and explains what remains unproved. Read the retained answers and the review method.
Read the beta evaluationEarlier results: the compiled-answer archive and the eleven-case diagnostic refresh.