Suki AI
The only ambient scribe with a real developer platform behind it, and the only one that published research saying the industry's own quality yardstick is broken. Independently validated at $1,223 incremental revenue per provider per month, and honest enough to report a 31% hallucination rate in its own peer-reviewed study.
PRICING
| Documentation tier (formerly Suki Compose) | ~$299/mo | per clinician, third-party estimate; ambient notes and dictation. Suki no longer names this tier publicly |
| Full assistant (formerly Suki Assistant) | ~$399/mo | per clinician, third-party estimate; adds deep EHR write-back, voice commands, coding and order staging |
| Enterprise | $350-$500+/mo | per provider, negotiated by contract on EHR depth, specialty mix and organisation size |
| Individual / small practice | Not sold self-serve | demo and pilot only; no posted price, no credit-card signup |
Every number here is third-party estimate. Suki publishes no pricing page, and both tier names above have been dropped from its own site. Treat the range as a procurement starting point, not a quote, and insist on a pilot in your own EHR build before signing.
checked 2026-08-17 · vendor pricing page
Element scores
Strengths
Two things separate Suki from the rest of the ambient category. First, the developer platform. Nobody else in clinical documentation ships REST APIs, a Web SDK, a headless React SDK, an iOS SDK, dictation and form-filling SDKs, WebSocket streaming and webhooks as a partner product, and it is doing real volume: ambient encounters across clinicians and partners grew 88% in six months to June 2026, embedded inside athenahealth, Zoom, WellSky, HealthEdge, Sevocity, MEDENT and Azalea Health. Second, the evidence. KLAS ran an independent ROI validation across FMOL Health, McLeod Health and Rush University System for Health and found a 35-65% reduction in after-hours documentation and an average $1,223 incremental revenue per provider per month, with McLeod reporting 27% less documentation time and a net $1,004 per provider per month. Austin Regional Clinic added an 18.5% cut in documentation time per encounter and a 97% engagement rate across 40 locations, which is the number that matters most, because the JAMA multisite work showed the benefit only lands for clinicians who actually use the tool in most visits. KLAS performance score is 93.2/100 with 95% saying they would buy again. Product breadth is genuine too: 100+ specialties, 80+ languages, voice commands, ambient order staging, chart Q&A, ICD-10, HCC, CPT and E/M coding.
Honest dings
Start with the number Suki published about itself. In its own peer-reviewed Frontiers in Artificial Intelligence study of 97 encounters and 388 paired reviews, hallucinations appeared in 31% of ambient notes against 20% of physician-authored ones, and physician notes still scored higher on accuracy, succinctness and internal consistency. That paper was written by Suki staff, which cuts both ways: it is a conflict of interest and it is more candour than the category usually offers. The vendor-neutral picture is harsher. A Veterans Health Administration study in Annals of Internal Medicine (April 2026) put 11 AI scribes against 18 human clinicians on five standardised primary-care visits and human notes scored higher in every case, with the gap widening under background noise, masks and non-native English speakers. So the clinician review step is not a formality, and Suki ships no human-QA layer and no automated hallucination gate to help. Commercially, there is no pricing page, no self-serve trial and no individual plan, so you cannot comparison-shop without a sales cycle. The voice-command interface is the differentiator and also the friction: reviewers consistently report it works well after clinicians memorise the phrasing, which means the thing you are paying extra for is the thing least likely to get adopted. Deep EHR integration covers exactly four systems, so eClinicalWorks, Veradigm, AdvancedMD and DrChrono practices fall back to copy-paste. And on the periodic table specifically, the model columns are closed: no model choice, no cheaper tier, no interpretability surface, no eval tooling for the buyer, and the MCP server is documentation only.
Sources (17) — 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).
- suki.ai — Official homepage: 400+ healthcare systems and partners, deep real-time integration with Epic, Oracle Health, athenahealth and MEDITECH, 100+ specialties, SOC 2 Type 2 and HIPAA, product scope across documentation, assisted revenue cycle and clinical reasoning (accessed 2026-08-17)
- suki.ai/partners — Official partner platform page: SDKs and APIs, ambient documentation, voice editing and personalization, 80 languages, diarization, problem-based charting, patient summaries, order staging, 41% documentation-time reduction and 81% practice-satisfaction claims, partner list (athenahealth, Zoom, WellSky, HealthEdge, Sevocity, Azalea Health, MEDENT) (accessed 2026-08-17)
- developer.suki.ai/llms.txt — Official developer documentation index: Ambient, Form filling, Dictation and Patient Summary products, GET /ws/stream and GET /ws/transcribe WebSocket boundaries, partner JWT/JWKS token-exchange auth, Web SDK, Headless Web SDK, Mobile SDK, LOINC note sections, agent-skills index (accessed 2026-08-17)
- developer.suki.ai/documentation/references/mcp — Official: Suki ships an MCP server at developer.suki.ai/mcp with search and get-page tools, and states explicitly that it does not execute Suki API calls on your behalf. Evidence for the Protocols score being documentation-only (accessed 2026-08-17)
- suki.ai/blog/two-years-in-suki-developer-platform-c… — Official 10/06/2026: modular platform history, pre-built UI versus headless/API paths, partner integrations compressed from six months to six weeks, ambient encounters up 88% in six months, MCP and agentic-workflow support claim (accessed 2026-08-17)
- pmc.ncbi.nlm.nih.gov/articles/PMC12586549 — Primary quality evidence, peer-reviewed. Palm et al., 'Assessing the quality of AI-generated clinical notes: validated evaluation of a large language model ambient scribe', Front Artif Intell 22/10/2025, doi:10.3389/frai.2025.1691499 — 97 encounters, 194 notes, 388 paired reviews across five specialties; Gold 4.25/5 vs Ambient 4.20/5 (p=0.04); ambient better on thoroughness and organisation, worse on accuracy, succinctness and internal consistency; hallucinations in 31% of ambient vs 20% of gold notes (p=0.01); reviewers preferred ambient notes 47% vs 39%. Conflict of interest: several authors are Suki AI staff (accessed 2026-08-17)
- fiercehealthcare.com/ai-and-machine-learning/rush-m… — Independent (Fierce Healthcare) 17/02/2026 on the KLAS validation: 35-65% reduction in after-hours documentation across three health systems, average $1,223 incremental revenue per provider per month, McLeod 27% documentation-time reduction and $1,004 net per provider per month, rollout from 40 to 249 users; also cites the UCSF JAMA Network Open study of 1.2M encounters finding +1.81 RVUs/week for scribe adopters (accessed 2026-08-17)
- klasresearch.com/report/suki-roi-validations-2026-c… — Independent KLAS ROI Validation, 22/01/2026: FMOL Health, McLeod Health and Rush University System for Health; reduced documentation burden, improved E/M coding, financial gain, three-to-five-month pilots subsequently expanded (accessed 2026-08-17)
- suki.ai/press-releases/austin-regional-clinic-cuts-… — Official 23/07/2026: Austin Regional Clinic, 700,000 patients across 40 locations — 18.5% reduction in documentation time per encounter, $1,452 average annual improvement per provider from E/M coding accuracy, 97% engagement among onboarded clinicians at 5+ encounters per week (accessed 2026-08-17)
- fiercehealthcare.com/ai-and-machine-learning/ai-scr… — Independent (Fierce Healthcare) 27/07/2026: Suki researchers argue PDQI-9 is structurally unfit for LLM errors; 84 paired notes across four specialties; raters disagreed on what counts as a hallucination; CMO Kevin Wang calls for vendors to publish inter-rater reliability rather than treating quality as a black box (accessed 2026-08-17)
- suki.ai/whitepapers/one-score-nine-dimensions-zero-… — Official whitepaper: PDQI-9 was validated in 2012 on 9 inpatient notes from 3 charts; Suki tested it on 84 paired notes with two production-candidate LLM models; alternatives assessed (PDSQI-9, SCRIBE, FActScore, VeriFact, CREOLA) and none combine sentence-level error detection, inter-rater reliability and a release-decision procedure (accessed 2026-08-17)
- mednews.uw.edu/news/AI-scribes-lower-quality — Independent vendor-neutral counterweight, 17/04/2026: Reddy et al., Annals of Internal Medicine (Veterans Health Administration) — 11 AI scribe tools and 18 human clinicians generated notes from five standardised primary-care visits, 30 blinded raters scored with modified PDQI-9, human notes scored higher in every case across accuracy, thoroughness, usefulness, organisation and comprehensiveness (accessed 2026-08-17)
- techtarget.com/searchhealthit/feature/AI-scribe-not… — Independent 20/05/2026: JAMA Network Open study (University of Virginia and others) of 11 AI scribes under realistic conditions — background noise, face masks, non-native English speakers — found AI tools less consistent than clinicians and raised equity concerns; recommends AI notes be treated as drafts requiring review (accessed 2026-08-17)
- litmustools.com/review/suki — Independent review 02/07/2026, scored 83/100: no self-serve tier or open trial, reported $299 Compose and $399 Assistant per clinician per month, voice-command learning curve, no human-QA review layer, vendor 99.4% transcription-accuracy claim flagged as unaudited, EHR coverage Epic/Oracle Cerner/athenahealth/Meditech/Elation (accessed 2026-08-17)
- plaud.ai/blogs/articles/suki-ai-review — Third-party review 26/05/2026: KLAS performance score 93.2/100 with 95% would-buy-again, founded 2017 by Punit Soni, roughly $168M raised and ~$500M valuation after the 2024 Series D, EHR coverage gaps outside the big four, and Suki's own statement to an American Psychiatric Association inquiry that its speech recognition does not detect emotional or nonverbal cues (accessed 2026-08-17)
- deepcura.com/resources/suki-ai-review — Competitor-published review 19/03/2026 (treat pricing and comparisons as interested): confirms no choice of AI engine, no self-serve signup, no prior authorisation, fax management or payment collection; reports drug-recognition misses and formulaic notes after extended use; pricing $299-$399 with enterprise $350-$500+ (accessed 2026-08-17)
- suki.ai/blog/clinicians-first-how-suki-is-building-… — Official 18/05/2026: 41% reduction in time per note, 60% reduction in burnout feelings, 81% increase in patient satisfaction (Press Ganey-derived), adoption rate over 80%, Suki Clinical Network and Clinical Catalyst teams, pause-and-switch concurrent care (accessed 2026-08-17)