The clinical AI market has fragmented into distinct categories serving different users, different workflows and different geographies. Understanding which category each tool belongs to, and which categories are available to UK clinicians, is the prerequisite for making informed decisions about adoption.
Category 1: General-purpose AI applied to medicine
These are the large language models clinicians already use, not because they were designed for clinical workflows, but because they are powerful, accessible and convenient.
ChatGPT (OpenAI): the most widely used AI tool among clinicians globally. Three health-specific products launched across 2026: ChatGPT Health (consumer, 7 January, connects medical records and wellness apps, excluded from UK, EEA and Switzerland), ChatGPT for Healthcare (enterprise, 8 January, HIPAA-compliant, GPT-5 models, deploying at eight major US institutions), and ChatGPT for Clinicians (individual, 22 April, free for verified US clinicians). On 1 September, ChatGPT for Healthcare specifically gained a read-only Epic EHR connection and a nine-source Healthcare Public Data plugin spanning PubMed, DailyMed, RxNorm and openFDA among others, with eligible ChatGPT for Clinicians accounts gaining the public-data plugin without Epic access. OpenAI's next foundation model, GPT-6 Astra, subsequently shipped with a stronger HealthBench Professional score and genuine agentic capability, browsing, file search, code execution and computer use, covered directly below and in full in this cluster's dedicated Astra coverage. Strong at evidence synthesis, writing and administrative tasks. Not UK-regulated, not optimised for UK guidelines, and none of the four health-specific capabilities are currently available to individual UK clinicians. The Nature Medicine study (Ramaswamy et al., Mount Sinai, February 2026) found 52% undertriage of emergencies in the consumer product specifically.
Claude (Anthropic): strong reasoning, safety focus and document-analysis capabilities, good at reading and analysing clinical papers, discharge summaries and complex clinical scenarios. No health-specific product or medical device registration, built on a constitutional AI approach with built-in safety constraints. Anthropic has separately introduced healthcare connectors covering sources including PubMed, CMS information, ICD-10 and the NPI Registry, positioning Claude alongside OpenAI in the connector race this landscape increasingly turns on.
Gemini (Google): Google Health integration developing, with Med-PaLM lineage and strong multimodal capabilities, particularly promising for imaging analysis across radiology, dermatology and pathology. Google's health data infrastructure, Fitbit and Google Health among it, positions the company for future consumer health AI, though no standalone clinician product exists yet. Perplexity: citation-first search AI with strong source attribution, every claim linked to its source, making verification faster than with conversational AI models generally, good for rapid evidence queries though not health-specific and carrying no regulatory status.
The arrival of GPT-6 Astra
Worth its own entry because it is a different kind of development from any single product launch above: a new underlying foundation model rather than a new healthcare product. GPT-6 Astra scored 63.4 on the length-adjusted HealthBench Professional evaluation against 60.5 for the prior GPT-5.6 Sol, with further improvement on the harder HealthBench variant, and carries an approximately 1.05 million token context window with genuine tool use, web search, file search, code execution, computer use, MCP and structured outputs.
The correct way to read this for the landscape as a whole: the medical-knowledge improvement is real and incremental, not transformative, and the more consequential change is agentic capability, the shift from a model that answers a question to one that can gather information, apply instructions, operate tools and produce a structured multistep result. This has direct implications across every category on this page, since a stronger underlying model raises the capability ceiling for general-purpose tools, specialist platforms built on top of frontier models, and health-tech products alike, without itself resolving any of the source-control, jurisdiction, provenance or regulatory questions this whole landscape turns on. OpenAI has not stated that Astra already powers ChatGPT for Healthcare or the Epic integration specifically, a distinction this cluster's dedicated Astra hub article and the OpenAI healthcare-strategy pillar both treat as an open question rather than an assumption.
Category 2: Clinician-facing clinical AI, purpose-built
Tools designed specifically for clinical use, built from the ground up to serve clinicians rather than adapted from general-purpose models.
OpenEvidence: the largest clinician-facing platform globally, a $12 billion valuation following a $250 million Series D in January 2026, approximately 15 million clinical consultations monthly from verified clinicians worldwide, free and ad-funded with pharmaceutical advertising during loading screens at reported CPMs of $70 to $150-plus. Embedded into Mount Sinai's Epic EHR in March 2026, placing AI clinical decision support inside the physician's primary workflow, with a Sutter Health Epic collaboration since February 2026 and patient-aware clinical-intelligence work underway with Cedars-Sinai. Uses proprietary models rather than ChatGPT, HIPAA-compliant and SOC 2 Type II certified, strongest for US evidence queries and peer-reviewed literature synthesis, US-centric in guideline coverage but globally accessible.
iatroX: UK-focused clinical AI platform, MHRA-registered, UKCA-marked Class I medical device, free. Ask iatroX retrieves and synthesises NICE guidelines, CKS summaries, peer-reviewed literature and SmPC data, grounded in the authoritative sources UK clinicians actually use, alongside 15-plus adaptive exam Q-banks covering all major UK exams, iatroX Simulations spanning eighteen examination-specific clinical performance tracks, 80-plus clinical calculators with editorial content and guideline references, and CPD documentation tools. The only platform combining clinical AI, exam preparation, clinical performance simulation, calculators and CPD in a single regulated device, purpose-built for UK clinical practice from foundation year to consultant.
Medwise AI: UK enterprise clinical AI with NHS Trust deployments, integrating local Trust policies, formularies and antimicrobial guidelines alongside national NICE guidelines, enterprise-licensed rather than available to individual clinicians, strongest for institutional workflow integration. Glass Health: differential-diagnosis generation from clinical presentations, US-focused, VC-backed, a narrower scope executed with focus. DxGPT: diagnostic AI from Foundation 29 focused specifically on rare diseases, not-for-profit.
Category 3: AI scribes and documentation
A distinct category frequently confused with clinical AI but serving a fundamentally different purpose. Scribes capture and document consultations; they do not answer clinical questions.
Heidi Health: custom templates, letter writing, clinical note generation, a zero data retention privacy model, popular in UK and Australian primary care. Tortus AI: UK-focused AI scribe with NHS deployments across more than 3,500 practices, ambient listening with structured note generation writing back directly into EMIS and SystmOne. Tandem Health: a European AI scribe reaching more than 200,000 NHS clinicians via an Accurx Scribe partnership, CE-marked and MHRA-registered. Nabla, Freed AI, Nuance DAX and Dragon Copilot, and Abridge round out the category, each with its own specific privacy model, integration depth and geographic focus.
Category 4: Enterprise clinical AI
Tools requiring institutional procurement, IT infrastructure and enterprise agreements, not available to individual clinicians.
ChatGPT for Healthcare (OpenAI): hospital-wide deployment, HIPAA-compliant with business associate agreements and customer-managed encryption, GPT-5 models at launch, already at eight major US institutions with citations from peer-reviewed studies, enterprise pricing. As of 1 September 2026, this specific product also carries a read-only Epic EHR connection, retrieving clinical notes, medications, conditions, encounters and laboratory results the individual clinician already holds permission to access, with UCSF Health named as the pilot organisation, alongside the nine-source Healthcare Public Data plugin. OpenAI reports physicians rated 99.1% of 4,363 evaluated responses safe across 27 EHR-related use cases, a self-reported figure worth reading against the same independent-evaluation caution the Nature Medicine study raises for ChatGPT Health elsewhere in this landscape. Whether and when this specific product gains GPT-6 Astra as its underlying model has not been confirmed by OpenAI.
Microsoft Copilot for Healthcare: enterprise only, six-figure annual commitments, requiring the Microsoft ecosystem, combining Dragon Copilot's clinical-documentation heritage with broader workflow automation. Doximity DoxGPT: available to US physicians through the Doximity professional network, with PeerCheck citation verification checking AI-generated references against real publications.
What a UK clinician can actually use today
Available to individual UK clinicians: iatroX, free and MHRA-registered, clinical AI plus exams plus simulation plus calculators plus CPD; Heidi Health, Tortus AI and Tandem via Accurx as AI scribes; standard ChatGPT, Claude and Gemini as general-purpose tools with appropriate safety caveats; and OpenEvidence, free and globally accessible though US-evidence-centric.
Not available to individual UK clinicians: ChatGPT Health, geo-blocked from the UK; ChatGPT for Healthcare, enterprise-only with no announced UK deployment, its 1 September Epic and public-data additions and any future Astra integration included; ChatGPT for Clinicians, appearing US-first; Microsoft Copilot for Healthcare, enterprise-only; and Doximity DoxGPT, restricted to US physicians. Available at institutional level only: Medwise AI, enterprise, NHS Trusts.
Where the market is heading
Convergence: OpenAI's three-tier stack, consumer through individual clinician through enterprise, is being replicated across the industry, with Claude for Healthcare and Gemini for Healthcare plausible within twelve to eighteen months, and Anthropic's own healthcare connectors already a live signal of this pattern. The frontier-model layer is converging too, as GPT-6 Astra's benchmark gains illustrate: incremental medical-knowledge improvement, larger agentic capability improvement, a pattern likely to repeat across every major provider's next model generation.
Regulation is coming: the MHRA's National Commission into the Regulation of AI in Healthcare is developing recommendations, and the UK will likely formalise requirements for clinical AI tools to hold medical device registration. Platforms already registered, iatroX with MHRA registration and Tandem with CE marking and MHRA compliance among them, hold a regulatory head start that takes years and substantial investment to replicate, a barrier new entrants including OpenAI will face regardless of how capable their underlying model becomes.
Guideline localisation will be the differentiator: as foundation model capabilities converge and strengthen together, the technical gap between tools narrows further, and the differentiator shifts even more decisively from does it know medicine to does it know my medicine. A more capable underlying model like Astra raises what every product built on top of it can plausibly do; it does not itself supply UK guidelines, UK licensing status or NHS-system integration, meaning the localisation gap this landscape describes throughout is, if anything, becoming more valuable to hold rather than less.
Workflow integration is the endgame: the winning tools will be embedded in clinical workflows, inside the EHR, triggered by clinical events, surfacing the right information at the right moment. Mount Sinai embedding OpenEvidence in Epic was an early signal; ChatGPT for Healthcare's own 1 September Epic connection is the clearest confirmation yet that this prediction was correct, with the enterprise EHR-integration race now genuinely underway between the two companies, and a stronger underlying model like Astra plausibly deepens rather than replaces that same integration race. iatroX's integration across exam preparation, clinical simulation, clinical calculators, clinical AI and CPD creates a different kind of workflow integration, spanning the learning-to-practice continuum rather than the single consultation.
iatroX's unique position
The only platform combining clinical AI, exam preparation, clinical performance simulation and CPD documentation within an MHRA-registered medical device, available in the UK, free, purpose-built for UK clinical practice. No other tool in any category, general-purpose, clinician-facing, scribe or enterprise, covers this breadth within a single regulated platform, and neither the 1 September announcement's enterprise EHR integration nor the arrival of a more capable underlying model like GPT-6 Astra changes that specific combination for a UK clinician today.
Frequently asked questions
Does GPT-6 Astra change which category a UK clinician should look to first?
Not directly: Astra is a foundation model rather than a UK-available healthcare product, and until it is confirmed to power a UK-deployable product with appropriate regulatory status, the practical landscape for UK clinicians described in this article remains unchanged.
Which category does the 1 September Epic integration actually belong to?
Category four, enterprise clinical AI, since it requires institutional Epic deployment and organisational configuration rather than anything an individual clinician's account can access alone.
Should UK clinicians expect this landscape to change again soon?
Very likely: this article names convergence, regulation and workflow integration as the clearest ongoing trends, and both OpenAI's product moves and the arrival of a stronger underlying model are each evidence those predictions are already playing out, with further competitor responses plausible within the timeframes this landscape sets out.
Try iatroX, clinical AI, exam prep, simulation and CPD in one MHRA-registered platform →