OpenAI has built a consumer-to-professional-to-enterprise healthcare stack in eight months, and no company has attempted this pace of healthcare AI product deployment before. Understanding the strategy, and its gaps, matters for every clinician evaluating which tools to adopt, and this page tracks the complete, dated picture across all three products, and now the foundation model underneath them, as OpenAI's healthcare strategy continues to develop.
The three-tier stack
Tier one, consumer: ChatGPT Health, launched 7 January 2026. Patients manage their own health, connecting medical records via US EHR/FHIR, Apple Health, and wellness apps including MyFitnessPal, Peloton, Function, AllTrails, Instacart and Weight Watchers. Users ask wellness questions, get plain-language lab explanations, and prepare for appointments. OpenAI reports approximately 40 million daily US health queries, developed with input from more than 260 physicians across 60 countries and over 600,000 feedback interactions across two years. Health chats are encrypted, isolated from general ChatGPT conversations, and not used for foundation-model training. UK, EEA and Switzerland are explicitly excluded, and medical-record integrations are US-only.
Tier two, individual clinician: ChatGPT for Clinicians, launched 22 April 2026, free for verified individual doctors, nurse practitioners, physician assistants and pharmacists, led by Nate Gross, OpenAI's Head of Health. This product fills the gap between the consumer tool, where patients self-triage with documented safety limitations covered below, and the enterprise tool, which requires institutional deployment and enterprise pricing individual clinicians cannot access. Verification requires an NPI number, a US-specific identifier UK clinicians do not hold, making the product US-first in practice.
Tier three, enterprise: ChatGPT for Healthcare, launched 8 January 2026 for hospital-wide deployment. HIPAA-compliant with customer-managed encryption keys, business associate agreements and institutional policy integration, powered by GPT-5 models at launch and evaluated via HealthBench, a physician-written rubric prioritising safety behaviours, and GDPval. Already deployed at AdventHealth, HCA Healthcare, Boston Children's Hospital, Cedars-Sinai, Memorial Sloan Kettering, Stanford Medicine Children's Health and UCSF, with citations from peer-reviewed studies and custom enterprise pricing.
What changed on 1 September 2026
ChatGPT for Healthcare specifically, not the consumer ChatGPT Health product, gained two new capabilities. Authorised Epic patient context: ChatGPT can retrieve information the individual clinician already holds permission to access within Epic, clinical notes, medications, conditions, encounters and laboratory results, oriented around practical questions like what has changed since the last appointment and which recent results need review. UCSF Health, already a ChatGPT for Healthcare customer per the deployment list above, is named specifically as the pilot organisation for this Epic integration. The connection is explicitly read-only, unable to amend records, place orders, send patient messages or override existing Epic permissions, built on Epic's FHIR environment with read scopes for medications, observations, allergies, diagnostic reports, encounters and clinical documents.
The Healthcare Public Data plugin connects ChatGPT and Codex to nine official sources: PubMed, ClinicalTrials.gov, DailyMed, RxNorm, openFDA, CMS Coverage, CMS Open Data, Medicare Care Compare and the NPI Registry, each answering a genuinely different question, mapped individually in this cluster's dedicated source-by-source explainer.
GPT-6 Astra: the new foundation model underneath
A further, distinct layer of this announcement deserves its own dated entry, because it is easy and wrong to blur together with the product news above. GPT-6 Astra, OpenAI's next foundation model, scored 63.4 on the length-adjusted HealthBench Professional evaluation, against 60.5 for the prior GPT-5.6 Sol, and improved further on the harder HealthBench variant. It carries an approximately 1.05 million token context window and supports web search, file search, code execution, computer use, MCP and structured outputs.
This is a genuinely different kind of announcement from the January and September product news above: Astra is a model, not a healthcare product. OpenAI has not stated that Astra already powers ChatGPT for Healthcare or the Epic integration, and this page treats that as an open question rather than an assumption, exactly the distinction this cluster's dedicated Astra hub article works through in full. The honest framing: Astra could strengthen eligible healthcare workflows as access expands, a materially different and more cautious claim than asserting the existing products already run on it.
The larger significance of Astra for healthcare is not a dramatic jump in medical knowledge, the HealthBench improvement is real and incremental rather than transformative, it is the combination of stronger reasoning, very long context and genuine agentic capability, web browsing, file search, code execution and computer use together, that makes multistep healthcare workflows considerably more plausible than a chat-only model ever could. This cluster's dedicated coverage of GPT-6 Astra for healthcare, linked throughout the site's Simulations, evidence-synthesis and health-tech content, covers this in full.
OpenAI reports physicians evaluated 27 EHR-related use cases
Pre-visit review, clinical timelines, medication review and handover summaries among them, and rated 99.1% of 4,363 responses safe, alongside a separate finding of more than 93% of responses rated good or better for accuracy across each of five tested connected data sources specifically, not all nine. This cluster's dedicated analysis of that specific figure works through exactly what it does and does not establish.
The technical foundation and its limits
OpenAI's internal benchmarks claim GPT-5.2 achieves 100% diagnostic accuracy on some standardised scenarios and 98.7% on healthcare-specific tests, built with the same 260-plus-physician, 600,000-plus-feedback-interaction process underlying ChatGPT Health. Internal benchmarks and independent evaluation tell different stories, however. The Nature Medicine study (Ramaswamy et al., Mount Sinai, February 2026) independently evaluated ChatGPT Health specifically and found 52% undertriage of gold-standard emergencies, 35% overtriage of non-urgent presentations, significant anchoring bias when social context minimised symptoms (odds ratio 11.7, 95% confidence interval 3.7 to 36.6), and inconsistent suicide-crisis safeguard activation.
The gap between internal benchmarks and independent evaluation is a consistent pattern across this category, not unique to OpenAI, and it is precisely why independent evaluation and regulatory frameworks matter, a manufacturer's own safety claim is not the same as an independent regulator or research team verifying it. This history is worth holding in mind directly when weighing any of OpenAI's own reported safety figures, including the 1 September Epic integration's 99.1% figure and Astra's HealthBench score alike, both self-reported and neither yet independently replicated.
The competitor landscape
The market OpenAI is entering is not empty, and each competitor occupies a distinct position. OpenEvidence, the largest clinician-facing clinical AI platform globally, reached a $12 billion valuation following a $250 million Series D in January 2026, processing approximately 15 million clinical consultations monthly, free and ad-funded with pharmaceutical advertising during loading screens at reported CPMs of $70 to $150-plus. It was embedded into Mount Sinai's Epic EHR in March 2026, the first enterprise-wide deployment across an entire care team, with a Sutter Health Epic collaboration announced in February 2026 and patient-aware clinical-intelligence work underway with Cedars-Sinai. It uses proprietary models rather than ChatGPT, is HIPAA-compliant and SOC 2 Type II certified, and remains US-evidence-centric while globally accessible.
Epic itself is not a passive record waiting for external AI: its own Art environment offers chart summarisation, draft documentation and, through Ask Art, natural-language patient questions answered with citations and visualisations, without manual chart searching. Medwise AI is UK enterprise clinical AI with NHS Trust deployments integrating local policy and formulary alongside national guidelines, enterprise-licensed rather than available to individual clinicians. Glass Health offers differential-diagnosis generation from clinical presentations, US-focused and VC-backed with a narrower, focused scope. Doximity DoxGPT is available to US physicians through the Doximity network, with PeerCheck citation verification integrated into a platform clinicians already use professionally. Microsoft Copilot for Healthcare is enterprise-only, requiring six-figure annual commitments and Microsoft ecosystem integration, building on Dragon Copilot's clinical-documentation heritage.
The UK gap, challenge and opportunity
None of OpenAI's three products, nor GPT-6 Astra itself, are currently available to individual UK clinicians in their intended clinical form. The UK regulatory environment, MHRA medical device classification requirements, UK GDPR health data protections, and the MHRA's National Commission developing AI healthcare recommendations, creates barriers OpenAI has not yet addressed. The 1 September nine-source data stack does nothing to close this gap, and neither does Astra's underlying model capability: PubMed and ClinicalTrials.gov are internationally relevant, while DailyMed, RxNorm, openFDA, CMS Coverage, CMS Open Data, Medicare Care Compare and the NPI Registry are specifically rooted in American healthcare and regulatory infrastructure, with no NICE, CKS, SIGN, MHRA information, UK SmPCs on emc or local NHS pathway equivalent announced, and no announced EMIS or SystmOne integration comparable to Epic's.
This creates a genuine paradox: the UK market is underserved by the world's largest AI company, while being well served by purpose-built UK-native tools that have already navigated the regulatory pathway, UKCA marking, MHRA registration, DCB 0129 clinical safety governance, any entrant including OpenAI would need to complete before UK clinical deployment. A stronger underlying foundation model does not shortcut this pathway; it simply raises the capability ceiling of whatever eventually reaches it. When ChatGPT for Clinicians or ChatGPT for Healthcare eventually reach the UK in a clinically deployable form, iatroX will already hold regulatory status, UK guideline depth and clinical trust that takes years to build.
What this means for the market
Free clinician-level AI from the world's largest AI company compresses prices and raises expectations across the entire market, and purpose-built tools must differentiate on dimensions a general-purpose model cannot easily replicate. Guideline specificity: NICE NG28 is not ADA guidance, and dose, threshold and management-pathway differences between jurisdictions are clinically significant, requiring local guideline layers global models do not build in. Regulatory status: MHRA and UKCA registration is a trust signal for clinicians, employers and increasingly regulators. And workflow integration beyond chat: exam preparation, clinical calculators and CPD documentation are capabilities ChatGPT does not offer, creating daily touchpoints a chat-only clinical AI cannot replicate, however capable its underlying model becomes.
The dated timeline
7 January 2026: ChatGPT Health launches for consumers, UK/EEA/Switzerland excluded. 8 January 2026: ChatGPT for Healthcare launches for enterprise deployment. 22 April 2026: ChatGPT for Clinicians launches for verified US clinicians. 1 September 2026: ChatGPT for Healthcare gains Epic EHR integration (UCSF Health pilot) and the Healthcare Public Data plugin; eligible ChatGPT for Clinicians accounts gain the public-data connectors without Epic access. GPT-6 Astra ships as OpenAI's next foundation model, with its healthcare-specific integration into the existing product stack not yet confirmed by OpenAI, covered in full in this cluster's dedicated Astra hub article.
Frequently asked questions
Does ChatGPT Health now connect to Epic?
No: the Epic integration belongs specifically to ChatGPT for Healthcare, the enterprise product, not to the consumer-facing ChatGPT Health product.
Does GPT-6 Astra power ChatGPT for Healthcare and the Epic integration?
OpenAI has not stated this, and this page treats it as an open question rather than a confirmed architecture; the two announcements may reasonably be discussed together as a direction of travel without being represented as the same confirmed technical system.
Has independent research validated any of OpenAI's healthcare safety claims?
The Nature Medicine study by Ramaswamy and colleagues at Mount Sinai independently evaluated ChatGPT Health specifically and found meaningfully different results from OpenAI's own internal benchmarks; no comparable independent evaluation of the 1 September Epic integration's 99.1% figure or of Astra's HealthBench score has been published as of this update.
Will this page be updated as OpenAI's healthcare strategy develops further?
Yes: this remains the tracking page for OpenAI's complete healthcare product line and its underlying foundation models, updated with dated entries as further launches, integrations or evaluations are announced.