Omi Medical v1.1

Medical speech-to-text, batch and live.

Transcribe consultations, dictation and calls with medical vocabulary, speaker labels and timestamps. Version 1.1 made 21% fewer medical-term errors than v1 on our live-transcription benchmark.

Live transcriptionListening
Draft · Speaker A

start amoxicillin five hundred milligrams and repeat hba1c at review

Confirmed · 00:04

Start amoxicillin 500 mg and repeat HbA1c at review.

speaker labels onmedical vocabulary onrealtime
Flagship API

Omi Medical v1.1

Our latest medical transcription release, for recorded audio and live conversations. Keep using API model ID omi-medical-1.

  • DeploymentHosted API or supported private cloud
  • IncludesBatch, realtime, vocabulary, speakers and timestamps
  • LanguagesEnglish benchmarked medically; seven more available for testing
#1 / 20batch board · US, October 2026
0.87%medical WER
96.81%dose / unit matches · 91/94

Batch: 131 ms median response (222 ms p95) on short clips. Realtime: final transcript 86 ms median after client-signalled speech end (169 ms p95). Every API measured in the US (Ohio), October 2026. See accuracy and speed →

New · Fast batch

Omi Medical-1 Flash

Medical transcription for recorded consultations, dictation and archives. $0.18 per audio-hour, 38% less than our flagship batch rate.

  • DeploymentHosted API, US processing
  • ModeAsynchronous jobs, transcript text
  • LanguageEnglish
#2 / 20batch board · tie-break on dose / unit matches
1.04%medical WER
96.81%dose / unit matches · 91/94

A 1-hour recording in 3.89 seconds median on a dedicated RTX PRO 6000, six runs with one request at a time. GPU benchmark; API upload and queueing add time. See the results →

Need an open model? Omi Medical Edge runs on Apple Silicon, NVIDIA GPUs and CPUs. Explore the model and runtime, or see the docs for API evaluation.

New in v1.1

Safeguards for real conversations.

Fewer false starts

New guards help prevent stray text from digital silence and tiny audio fragments at the end of a turn.

Speech under hum

Improved speech detection helps recover speech masked by electrical hum. Only speech detection is filtered; the recognizer hears the original audio.

Still review

Mumbled or heavily masked speech can still be misheard. Review medical transcripts before use.

Hosted boundary

Product facts you can verify.

These statements apply to Omi's hosted speech API. Open and customer-managed deployments run inside the environment you control.

Open the Trust Center →
Self-serve BAA and DPAExecute both in the console on every hosted API tier, including Builder.Read the BAA →
Regional processingEU-only endpoint available. US processing and upcoming global routing are described in the DPA.Read the DPA →
Zero retention on sync requestsAudio is deleted after processing; transcripts aren’t stored. For async results, you choose the retention window: 1–72 hours.See the data flow →
No model trainingCustomer audio and transcripts are never used to train Omi models.View subprocessors →
Also available through Telnyx

Telnyx offers omi-health/omi-med-stt-v1 as a speech-to-text engine for clinical intake, medication review, dictation and voice assistants.

Read the Telnyx release note →