Glass Health

Glass Health is a clinical AI platform built for physicians, nurse practitioners, and physician assistants who want to spend less time on documentation and more time on patient care. Founded in 2021 by Dereck Paul, MD, an internal medicine physician who trained at UCSF and Brigham and Women's Hospital, and Graham Ramsey, an engineer with healthcare product experience at Modern Fertility and Plushcare, the company started as a physician notebook for organizing diagnostic approaches and pivoted into generative AI in 2023. As of May 2026, Glass Health has shipped five major platform versions and a public Developer API, accumulating more than 131,000 monthly active users and partnerships with enterprise health systems.

The platform combines two functions that most clinical AI tools separate: ambient scribing during the patient encounter and structured clinical decision support before, during, and after the visit. Core features include differential diagnosis generation with ranked "most likely," "expanded," and "can't miss" categories; a Consult mode that answers clinical questions with inline citations from 38 million peer-reviewed articles and 154,000 FDA-approved drug records; an Assessment and Plan generator that produces problem-oriented documentation; chart summarization from EHR data; and a full suite of documentation templates (H&P notes, daily progress notes, discharge summaries, patient handouts). EHR integration with Epic, eClinicalWorks, athenahealth, and Elation runs over SMART on FHIR, pulling patient context in and pushing completed notes back without copy-paste. A public API lets healthcare developers embed Glass's clinical reasoning into third-party applications.

What Glass Health actually is in May 2026

Glass Health's current product, Glass 5.5, is best understood as a clinical reasoning layer that wraps around the physician encounter rather than sitting beside it. When a clinician opens a patient file, they can activate ambient mode: Glass listens as the history unfolds, continuously refines a differential diagnosis in the sidebar, flags "can't miss" diagnoses (pulmonary embolism, aortic dissection, sepsis), and suggests follow-up history questions and physical exam components in real time. When the encounter ends, Glass generates a structured SOAP note or A&P document that the clinician edits and pushes directly to the EHR.

Outside of encounters, clinicians use the Consult mode as a clinical search engine: ask a clinical question in natural language and receive an evidence-grounded answer with citations to NEJM, UpToDate, and primary literature. The Workspace feature functions as a tabbed clinical notebook where each patient file holds the encounter transcript, generated differential, documentation drafts, and version history in one place. The Deep Reasoning mode, added in mid-2025, allocates the platform's maximum compute budget to complex cases, producing longer evidence synthesis and more nuanced differential reasoning at the cost of slightly slower response time.

The February 2025 launch of Glass 4.0 represented the clearest inflection point in the product's maturity. Before 4.0, Glass operated on a narrower medical knowledge base and had limited multi-turn conversational capability. The 4.0 release added continuous chat, advanced reasoning, and what the company described as a 275x expansion in medical literature coverage, moving from a differential-diagnosis specialist to a more comprehensive clinical AI. The December 2025 Glass 5.0 release followed with EHR connectivity, patient file creation directly within Glass, and file upload for chart context, bringing the ambient and CDS workflows into a single unified product for the first time.

Where Glass Health sits versus OpenEvidence and Hippocratic AI

OpenEvidence is the most direct competitor on clinical decision support, but the two tools solve different problems through different architectures. OpenEvidence trains specialized medical language models on 35 million peer-reviewed publications, with official content partnerships from NEJM, JAMA, the Cochrane Library, NCCN, and Wiley. Its design enforces a no-hallucination policy: the system refuses to answer when evidence is insufficient rather than generating plausible-sounding but unsupported text. It is free for verified U.S. clinicians and funded by publisher partnerships rather than physician subscriptions. The tradeoff is workflow integration: OpenEvidence is a question-and-answer tool you open between tasks; it does not listen to your patient encounter, does not generate notes, and as of early 2026, does not push output into the EHR (a Sutter Health Epic pilot was underway but not broadly available).

Glass Health's advantage over OpenEvidence is the encounter-native workflow. The ambient scribing, real-time differential refinement, and EHR push/pull mean a clinician never leaves the patient context to consult a separate tool. OpenEvidence's advantage is evidence depth and citation rigor. A physician treating a straightforward acute presentation might prefer Glass for speed and workflow continuity; a physician managing a rare disease or complex multi-drug regimen might prefer OpenEvidence's literature anchoring. Many clinicians pair both.

"Glass Health uses simpler language and is much easier to read." - Farah Deshmukh MD MPH, Rethink Health MD Substack, 2024

Hippocratic AI operates in a different part of healthcare AI entirely and rarely competes with Glass Health directly. Where Glass is deployed at the clinician workstation, Hippocratic AI's Polaris architecture (a constellation of specialized models that supervise a main conversational agent) is built for patient-facing phone and video interactions: appointment preparation, medication adherence reminders, post-discharge follow-up, and health education scripts. Hippocratic AI explicitly avoids diagnostic tasks; its core value proposition is reducing healthcare staffing burden by handling repetitive patient communication that does not require a licensed clinician. A health system might use both: Glass for physician encounter support and Hippocratic AI for the patient outreach layer. They are not substitutes.

For documentation-only workflows, Abridge and Nabla are worth comparing. Both focus on ambient scribing without Glass's clinical decision support layer. Abridge has deeper Epic-native integration (it is embedded inside Epic rather than connecting via FHIR), which matters for institutions running Epic as the primary workflow surface. Nabla covers a broader range of clinical specialties and has a stronger European regulatory footprint. Suki AI sits in the same ambient documentation space, offering voice-driven note dictation integrated with EHR systems.

How AI actually works inside Glass Health

Glass Health does not run a commodity GPT wrapper. The platform uses a proprietary clinical AI model (Glass 5.5 as of early 2026) evaluated against nine clinical accuracy benchmarks spanning medical knowledge recall, multi-step diagnostic reasoning, hallucination detection, evidence appraisal, and structured documentation generation. According to the company's internal benchmarking, Glass 5.5 outperforms frontier models from OpenAI, Anthropic, and Google on these nine benchmarks when measured against established external clinical AI evaluation datasets using identical prompts. These claims are [unverified by independent third parties] as of the research date, though the methodology involves established external benchmark datasets graded against their official answer keys.

The retrieval architecture uses RAG (retrieval-augmented generation) against a curated clinical knowledge base. When a physician inputs a patient summary or asks a clinical question, Glass retrieves physician-validated clinical guidelines as context before generating a response. Founder Dereck Paul described this in the TechCrunch interview that introduced the platform: "Our LLM application retrieves physician-validated clinical guidelines as AI context at the time it generates outputs." The Consult feature's inline citations trace back to specific articles in the literature corpus, though the citation interface is less granular than OpenEvidence's point-to-source linking.

The ambient scribing layer processes real-time audio during patient encounters and generates a rolling differential diagnosis as the conversation unfolds, updating the ranked list as new history elements are disclosed. The system is trained to distinguish clinically significant statements from routine social conversation and to maintain HIPAA-compliant handling of PHI, which requires a Business Associate Agreement for covered entities.

"Glass Health has completely transformed the way I approach medical learning and clinical decision-making." - Venance09, App Store review, cited in Clinical AI Report, 2025

The friction points physicians keep raising

The most consistent criticism across reviews is that Glass Health performs well on focused single-system presentations but becomes less reliable on complex, multi-system cases where cross-specialty reasoning is needed. A patient with simultaneous cardiac, renal, and hepatic involvement creates a differential space that requires weighting probabilities across fields in ways that Glass's model sometimes handles less gracefully than a senior hospitalist would.

The pricing structure creates friction for solo and small-group practice physicians. The Starter plan at $20/month covers limited functionality; the Pro plan at $90/month unlocks the full ambient scribing and CDS stack; the Max plan at $200/month is required for EHR integration with Elation. For a physician coming from a free OpenEvidence account, the value proposition requires concrete productivity evidence before committing to the Pro tier.

EHR integration, while a genuine differentiator, is uneven across systems. Epic, eClinicalWorks, athenahealth, and Elation are supported, but physicians on Cerner, Meditech, or smaller EHRs have no native integration and must rely on manual copy-paste. The SMART on FHIR approach is standards-compliant but depends on the institution's EHR configuration, which varies.

Early reviewers flagged the absence of drug dosing information as a significant gap, though the Glass 4.0 update added 154,000 FDA drug records and expanded pharmacology coverage. The previous criticism was fair for versions 1-3; it applies less to current Glass.

Who Glass Health is for

Glass Health fits best for outpatient physicians and clinical trainees who move through high volumes of diverse presentations and need structured diagnostic scaffolding alongside fast documentation. Residents and fellows find particular value in the systematic differential generation, which reinforces evidence-based reasoning rather than pattern-matching intuition during the learning phase. Solo practitioners and small-group practices who cannot afford enterprise scribing software will find the Pro tier competitive with dedicated scribing services.

Healthcare technology companies building clinical AI products are a growing use case via the Developer API. The API exposes Glass's core clinical reasoning, including evidence Q&A, patient data summarization, differential diagnosis, treatment planning, and documentation generation as callable endpoints, enabling smaller teams to embed clinical-grade AI without training their own medical models.

Skip Glass Health when the primary need is pharmacology depth alone (UpToDate or Epocrates are more comprehensive drug references), when the institution requires a tool embedded natively inside Epic without external API calls (Abridge or DAX Copilot serve that need better), or when the clinical workflow is emergency medicine with sub-minute patient turnover (ambient scribing during fast-paced ED encounters is harder to operationalize). Patients cannot access Glass Health directly; it requires a clinical login and is not designed for consumer health queries.