Medmio AI Hits 99.2% Medical Coding Accuracy in First Ultrasound Benchmark

Tested for the first time on diagnostic ultrasound, Medmio’s new model achieved 99.2% accuracy – topping their previous model’s 98.1%.

SAN DIEGO, CA, UNITED STATES, August 11, 2026 /EINPresswire.com/ — Medmio’s newest CodeSight™ model coded diagnostic ultrasound reports for the first time and matched the codes a vascular practice billed 99.2% of the time – the highest accuracy Medmio has published. The result comes from a 1,287-study benchmark in which 254 held-out studies were scored once, at the end.

The figure tops the 98.1% Medmio reported in July, when its first public benchmark measured CodeSight on office-visit charts. An office-visit chart is a doctor’s written account of a patient visit. An ultrasound report is a technical record of an imaging exam – protocol, findings, laterality.

Both tasks are medical coding – turning a clinical note into the standardized diagnosis and procedure codes a claim is paid on. Practices pay for that work chart by chart, and miscoded charts come back as denials – leading to reworked claims, delayed revenue, and bills that can land on the patient.

“Model upgrades are the heartbeat of an AI product, and practices deserve to see each one measured in public,” said Levan Sulakvelidze, President and CEO of Medmio. “We shipped a stronger model, pointed it at a kind of note it had never coded, and it came back more accurate than the number we published a month ago. That is the pattern we intend to repeat: every new model, benchmarked against real billed claims, published either way.”

CodeSight is the AI coding engine inside Medmio’s platform for physician practices, combining several large language models with deterministic, policy-based rules; the platform also includes voice dictation, patient intake and mobile charge capture.

The benchmark follows the same protocol Medmio used in July: the scored studies were held out before development began, scored once at the end, and evaluated against a configuration fixed in advance. The full method and limitations are published with the result, and a manuscript is in preparation. A third benchmark, covering surgical operative notes – a substantially harder coding task – is underway and will be published on the same terms.

The full ultrasound benchmark is published at https://www.medmio.com/case-studies/codesight-ultrasound-benchmark/. The July office-visit benchmark remains available in full.

About Medmio
Medmio is an AI-powered platform for physician practices that brings autonomous medical coding, voice dictation, patient intake, and mobile charge capture into one system. Its CodeSight™ engine reads clinical notes and produces billing and diagnosis codes with reviewable confidence scores, and integrates with major EHRs via HL7 standards. Learn more at www.medmio.com.

Levan Sulakvelidze
Medmio, Inc.
press@medmio.com
Visit us on social media:
LinkedIn
Instagram
Facebook
YouTube
X
Other

Media gallery