Heidi
The most transparent clinical AI vendor on the market and the only one that publishes a EUR price — but it is walking away from the small practices that made it.
PRICING
| Free | Free | unlimited transcription, standard templates, limited Ask Heidi and Evidence; free forever, no card required |
| Clinician | $150/mo | $110/mo billed annually (EUR €80 / €60); 14-day free trial |
| Practice | $180/mo | per user, annual only (EUR €75); adds team controls, EHR integration add-on, coding, template sharing |
| Enterprise | Custom | SSO, service availability terms, custom hosting, dedicated customer success; 30 to 1000+ users |
Heidi prices in 19 currencies and the euro list is materially cheaper than the dollar list — €60/month annually against $110. Read the price in your own currency before budgeting. Note also that these figures follow a February 2026 restructure that raised the core tier and moved deeper EHR write-back behind Practice.
checked 2026-08-12 · vendor pricing page
Element scores
Strengths
Heidi publishes more about how it tests itself than anyone else in this category, and the detail is not marketing. Scribe evaluation runs seven parallel layers; Evidence is scored on 15 named dimensions where faithfulness, factfulness, harm-risk extent, harm-risk likelihood and adversarial security form a hard safety gate that must clear before anyone asks whether the answer is good. Every candidate runs blinded side-by-side against the current production version, and the pipeline feeds a post-market surveillance plan under EU MDR Article 83. It also owns its models rather than renting them: a fine-tuned NVIDIA Parakeet V2 ASR at 9.4% WER against 13% for the legacy vendor model, with latency down from ~3.0s to ~0.7s and transcription costs down 64%, plus a self-trained Evidence model that reached blinded parity with Sonnet 4.6 at a fraction of the size. Owning the weights is what lets Heidi host in-region for data residency and keep Evidence free. Independent King's College London testing published in the Journal of Dental Research ranked Heidi the most accurate of four commercial ASR systems tested (6.2% domain WER against 33.9% for the worst). Add the world-first ISO 42001 certification, 110 languages, 116 countries, a genuine free tier, free access for trainees, and a published euro price — the only vendor on this list that will tell a Belgian clinician what it costs without a sales call.
Honest dings
The commercial direction is moving away from the clinicians who built the company. Heidi restructured pricing in February 2026, raising the core tier and moving deeper EHR write-back behind Practice, and trade reporting says it has ceased support for its small-practice product with early adopters told it will be enterprise-only. The Series B is explicitly earmarked for hospital sales teams in the US, UK and Canada. If you are a two-doctor clinic, the vendor you pick today may not want you in eighteen months. The developer story has a matching hole: the API and Widget are closed to new integrations in Australia and New Zealand — its home market — which is a strange thing to find at the bottom of a developer-docs page. On the periodic table the gaps are protocols (no MCP, no A2A), multi-agent orchestration, and customer-facing fine-tuning or evaluation. And for all the published methodology, the independent efficacy evidence is thinner than Nabla's: no randomised trial, and the closest external note-quality study is a six-consultation UK general practice study scoring 84% accuracy with five inaccuracies judged likely to make a clinical difference and one hallucination. Adoption context matters too — a May 2026 survey of 1,535 Australian GPs found only 18.7% personally use any AI scribe, so the 2M-consults-a-week figure reflects intensity among adopters, not market saturation.
Sources (16) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- heidihealth.com/pricing — Vendor's live pricing configuration read directly on 12/08/2026 across 19 currencies. EUR: Clinician €80/month monthly or €60/month yearly (saving €240), Practice €75/user/month annual, Enterprise custom; USD equivalents $150 / $110 / $180. Euro locale list includes nl-BE and fr-BE (accessed 2026-08-12)
- heidihealth.com/resources/evaluation-framework — Official evaluation framework: seven parallel Scribe layers, 15 Evidence dimensions across safety/quality/performance tiers (faithfulness, factfulness, harm risk extent and likelihood, security, completeness, fully cited, source authoritativeness, latency, information density, response consistency), blinded side-by-side against production, post-market surveillance under EU MDR Article 83 (accessed 2026-08-12)
- heidihealth.com/en-au/blog/how-heidi-improved-asr-n… — Official engineering post 16/03/2026: fine-tuned NVIDIA Parakeet V2 0.6B TDT on 1,500 hours of clinical audio, non-curated WER 9.4% vs 11.5% base and 13% legacy vendor, medical-term F1 93.5%, latency ~3.0s to ~0.7s, 64% transcription OpEx reduction, 2.4M weekly consultations, 100M+ sessions (accessed 2026-08-12)
- heidihealth.com/en-us/blog/clinical-ai-model-fine-t… — Official CEO post 25/06/2026: in-house Evidence model reaches 49.9% blinded side-by-side win rate against Sonnet 4.6, trained via SFT on preference-filtered rollouts, on-policy self-distillation and DPO; safety checked against HealthBench Pro and in-house Medical QA; Evidence at 300,000+ questions/week, 3.5M since launch; states scope and timing limits plainly (accessed 2026-08-12)
- kclpure.kcl.ac.uk/portal/en/publications/transcript… — Independent peer-reviewed evidence. King's College London, Journal of Dental Research: 200 clinical summaries, 43,408 words, 6 hours of audio across 4 commercial ASR systems — Heidi-Health highest-performing commercial system (domain WER 6.2%, WER 5.4%) vs DragonAnywhere worst (33.9%); no hallucinations from Heidi; clinically significant errors persist across all systems (2-66%) (accessed 2026-08-12)
- iaria.org/conferences2026/filesAIHealth26/80086_AIH… — Independent small-scale study (Shetye et al., AIHealth 2026): free Heidi vs medical-student notes across six English GP consultations, 84% accuracy, 5 of 32 scored items inaccurate in a way that would make a clinical difference, one hallucination; authors conclude accuracy declines with case complexity and human review remains essential (accessed 2026-08-12)
- heidihealth.com/en-us/blog/iso-42001-certification — Official 28/01/2026: first ambient medical scribe certified to ISO/IEC 42001:2023 for AI management systems, alongside SOC 2 Type II, ISO 27001 and Cyber Essentials Plus recertification; three-year certification cycle with external audits (accessed 2026-08-12)
- heidihealth.com/en-us/ai-instructions — Official vendor-maintained fact sheet (updated June 2026): 2M+ consults/week, 116 countries, 110 languages, 200+ specialties, 3.1 billion minutes transcribed, founded 2021 in Melbourne, product line (Scribe, Evidence, Remote, Forms, Dictate, Tasks, Template Community, Medical Coding), regional data residency across AU/US/UK/CA/EU with no cross-region transfer (accessed 2026-08-12)
- heidihealth.com/developers/heidi-api/overview — Official developer documentation: Heidi Open API at registrar.api.heidihealth.com/api/v2, JWT authentication, Postman collection and Swagger definition, Widget and TypeScript SDK for React/Angular/Vue, FHIR-compatible — carries the standing notice that the API and Widget are no longer available for new integrations in Australia and New Zealand (accessed 2026-08-12)
- support.heidihealth.com/en/articles/14698472-heidi-… — Official help-centre article on Heidi Verify: entity-level extraction and checking of notes against transcript and context, links each entity back to its source span, suggested fixes, explicit limitation that it cannot catch errors originating in the transcript; limited beta on Practice and Enterprise only (accessed 2026-08-12)
- medicalrepublic.com.au/heidi-swims-upstreaminto-the… — Independent trade reporting (The Medical Republic, 07/07/2026): February 2026 pricing restructure, core tier rebranded Clinician at $150/user/month annual and deeper EHR write-back moved behind Practice at $180; support ceased for the small-practice product with early adopters told it is going enterprise-only; Beth Israel Lahey Health rollout; Series B earmarked for US/UK/Canada hospital sales (accessed 2026-08-12)
- healthservicesdaily.com.au/is-heidis-everyone-every… — Independent critical analysis (Health Services Daily, 20/05/2026): median 86% utilisation across Australian hospital deployments, but a HealthEd/Medical Republic survey of 1,535 Australian GPs found only 18.7% personally use an AI scribe, and a Nuffield Trust/RCGP survey of 2,108 UK GPs found just over a quarter use AI at all; BILH deployment across 6,000 physicians and 14 hospitals (accessed 2026-08-12)
- heidihealth.com/en-au/blog/heidi-series-b — Official 05/10/2025: $65M USD Series B led by Point72 Private Investments at a $465M valuation, ~$96.6M total raised; 73M consults over 18 months, 18M clinician hours returned; Monash Health, Queensland Health Children's Hospital, Yukon Government (accessed 2026-08-12)
- businessinsider.com/exclusive-pitch-deck-ai-scribe-… — Independent (Business Insider, 08/10/2025) corroboration of the $65M Series B at $465M post-money; founded 2019 as Oscer by Dr Thomas Kelly, pivoted to clinical documentation (accessed 2026-08-12)
- healthcareitnews.com/news/anz/cabrini-ed-care-wraps… — Independent (Healthcare IT News): Cabrini Health emergency department pilot — clinicians using Heidi in over half their consultations saved an average 31 minutes on clinical notes and completed ED care 24 minutes earlier; 97% wanted continued access; New Zealand endorsement as one of two AI providers safe for public health system trials (accessed 2026-08-12)
- heidihealth.com/en-au/blog/the-royal-childrens-hosp… — Vendor case study (treat as vendor-reported): Royal Children's Hospital Melbourne deployment across 350 specialist clinics; 2025 Impact Report claims average 59 minutes of capacity recovered per day, 81% reporting reduced cognitive load, 78% greater enjoyment at work (accessed 2026-08-12)