Medical speech-to-text benchmark
Batch and realtime accuracy, with measured response speed.
Explore the benchmark →Add medical speech-to-text to a prototype, a startup product or a healthcare AI platform. Version 1.1 made 21% fewer medical-term errors than v1 on our live-transcription benchmark.
For clinicians who code · Healthtech startups · Healthcare AI companies
“Start amoxicillin 500 mg and repeat HbA1c at review.”
Same 1,513 clinical recordings for every model. What's new in v1.1 →
Use the same medical speech layer across live conversations, recorded notes and local workflows.
Capture speakers, timing and the medical language needed for notes and downstream agents.
Give clinicians a faster input method inside EHR, dental, therapy and specialty workflows.
Feed confirmed medical text into triage, follow-up and care-navigation logic.
Run the open model in your own application when audio cannot leave the environment.
Use batch and live transcription with medical vocabulary, speakers and timestamps.
Deploy omi-medical-edge-1 on Apple Silicon, NVIDIA CUDA or CPU—or try its English hosted endpoint first.
Compare medical terms, drug names and doses. Then see how quickly the final transcript arrives.
Selected batch results. Every API timed from the US (Ohio), October 2026. Speed: clips under 30 seconds. Full comparison →
Realtime: 1.04% medical-term error, lowest of 12 tested. Final transcript 86 ms median after client-signalled speech end (169 ms p95), measured in the US. See realtime results →
Open benchmarks, published model results and reproducible evaluation.
Batch and realtime accuracy, with measured response speed.
Explore the benchmark →What changed in v1.1: realtime accuracy, speed and safeguards for difficult audio.
Read the model report →Fast English batch transcription in the US at $0.18 per audio-hour.
Read about Flash →Use the hosted API or download the open model.