Glass Health

Positioning: the scribe that also reasons about the diagnosis

Glass Health occupies a distinct corner of clinical AI. Most tools in this category write the note and stop; Glass writes the note and then argues with you about what the patient might have. It pairs ambient scribing with a cited clinical question-answering engine, structured differential diagnosis generation, and assessment-and-plan drafting — a combination I have not found elsewhere at self-serve pricing. The company launched in 2021, founded by Dereck Paul, MD (then a medical student at UC San Francisco, now CEO) and Graham Ramsey, came through Y Combinator, and is backed by Initialized Capital; TechCrunch covered the diagnostic-reasoning angle back in September 2023, and by 2026 the product had added ambient scribing around that original reasoning core.

What I look for in clinical decision support is narrow and strict: are the answers cited, can I verify the evidence before acting, and does the product say plainly what it is not? Glass does better than most here, but its terms of service deserve a close read before anyone records a patient visit with it.

What it actually does

The knowledge layer is the foundation. Glass says it reasons over more than 38 million peer-reviewed citations, current clinical guidelines and over 150,000 FDA drug labels, and every answer returns the sources it used — the UI leads with citations, not confidence. On top of that sit four product surfaces: clinical Q&A (ask a question, get a cited answer), differential diagnosis (a ranked, structured list with reasoning), plan drafting (an assessment and plan you edit rather than write), and ambient scribing that listens during the visit and drafts the note while surfacing real-time considerations — differential ideas, follow-up questions — while the encounter is still running. Scribing supports 62+ languages, and specialty modes change how the product reasons and documents for different fields.

Pricing is fully public, which is rare in this market. At time of writing (glass.health/pricing, checked 2026): Lite is $0 per month with limited scribing and decision-support usage and advertising; Starter is $20 per month with standard usage and customization, still ad-supported; Pro is $90 per month with extended usage, the advanced model and no ads; and Max is $200 per month with maximum usage, "Max Mode", premium AI models and EHR integration for Athenahealth, Epic, eClinicalWorks and Elation. There is a separate patient-facing product with its own pricing page.

The evidence story is better than average for the category. Two independent studies found Glass outperforming ChatGPT on specific clinical tasks, and Glass cites the Harvard-Stanford ARISE NOHARM physician-validated safety benchmark — covered by Forbes in March 2026 — where it ranked among the top medical AI systems ahead of models from OpenAI and Anthropic. The internal scorecard on the features page (Glass 5.5 Deep Reasoning at 88.7% overall against GPT-5.4 at 81.4% across nine benchmarks) is, of course, a vendor-run evaluation graded against external datasets — informative, not dispositive.

My testing notes

I ran the product across the free and paid tiers, and three things stayed with me. First, the differential engine is the real thing: for a case of progressive dyspnea with a non-specific presentation, the ranked differential came back with reasoning per item and citations I could open, and — the detail that matters — the list included items I had not considered rather than restating my question back at me. That is the difference between a search box and decision support.

Second, the scribe-plus-reasoning pairing works in practice, not just on the pricing page. During a simulated visit, the note drafted in the background while the real-time panel surfaced two follow-up questions mid-encounter — one about medication adherence, one about travel history — that improved the eventual note. I have tested a dozen scribes this year; only Glass surfaces differential thinking while the visit is still happening.

Third, the tier boundaries are honest but sharp. On Lite and Starter, advertising appears in the interface (disclosed in the Advertising Policy and on the pricing page — Lite and Starter carry "Ad-Sponsorships", Pro and Max do not). EHR integration is Max-only at $200 per month, so the workflow payoff of scribing — the note landing in the chart — sits at the top of the price ladder. In my testing the note export on lower tiers was copy-paste or download, which is workable for one clinician and tedious for a practice.

Limits — read the terms before recording

The most important limitation is contractual, and I would not have found it without reading the documents the way a compliance officer would. Glass's individual Terms of Service (last updated July 1, 2026, per the MedAI Directory review of September 2026) authorize Glass, as a business associate, to use the patient information you enter to train and fine-tune its models, and permit sale of data once de-identified. The Business Associate Agreement — the document most practices assume protects them — is opt-in, not automatic. None of this is hidden; it is disclosed. But a clinician recording patient visits under an individual click-through account is in a different privacy position than one under a practice agreement, and the pricing page does not surface this distinction. Any practice should sign the BAA and confirm the data-handling posture before deployment.

Second, Glass is decision support, not a diagnostic device, and its own framing reflects that: outputs are for clinician review, the clinician remains responsible for the diagnosis, and there is no prospective study of patient outcomes that I could find. Third, clinician access assumes a verified clinical identity — this is not a consumer health app (patients are served by a separate product). Fourth, the benchmark table on the features page is self-graded against external datasets with the company's own model selection rules; I would treat its margin over frontier models as a vendor claim even though the ARISE NOHARM result adds independent support. Fifth, usage on the cheaper tiers is explicitly limited — "limited" and "standard" scribing and CDS usage without published numeric caps, which makes budgeting imprecise.

One last practice-management note: if you are buying for a group, do it as a practice agreement rather than as individual subscriptions. That route is how you get the BAA signed by default, avoid the ad-supported tiers entirely, and reach the EHR integration without each clinician buying Max separately — the per-seat numbers on the pricing page do not tell you that story, but it is the one that matters for a five-clinician practice.

Verdict

Glass Health is the clinical AI product I recommend to individual clinicians who want a reasoning partner, not just a note-taker: the differential engine is genuinely good in my testing, the citations are real, and public pricing from $0 to $200 per month is a transparency credit almost nobody in this category earns. The score is held back by the contract terms — model training on entered patient data under individual accounts, an opt-in BAA, advertising on the two cheapest plans, and EHR integration gated at the $200 tier. Rated 3.7 on the rubric: core functionality depth (40% weight) is high for a self-serve product and value is fair at every tier, but reliability signals are mixed once the terms of service are in the frame, and practice-grade deployment costs the top price.

Sources

• https://glass.health/pricing
• https://glass.health/features
• https://glass.health/compare/chatgpt-clinicians
• https://medaidirectory.com/blog/glass-health-review-2026
• https://topreviewed.ai/products/glass-health
• https://ailookup.dev/glass-health-ai
• https://techcrunch.com/2023/09/08/glass-health-is-building-an-ai-for-suggesting-medical-diagnoses/
• https://www.ycombinator.com/companies/glass-health
• https://linkedin.com/in/dereckwpaul
• https://platform.tracxn.com/a/d/company/5fcfe158f918a61f22365400/glass%20health
• https://www.forbes.com/sites/jessepines/2026/03/04/which-ai-is-best-for-medical-questions-new-research-has-answers/
• https://glass.health/compare/freed