Microsoft published language coverage, latency, pricing and access paths for the three speech models.
Microsoft releases streaming speech models with 100 ms partials
2026-10-01Verified announcementTRIAL
Microsoft released MAI-Transcribe-2-Streaming plus MAI-Voice-2.1 and MAI-Voice-2.1-Flash. The transcription model supports 60 languages, continuous language detection and initial partials just over 100 ms. Voice covers 23 languages and 26 locales; Flash targets 45 seconds of audio in 150 ms. The models are callable through Microsoft Foundry and public playgrounds.
Availability: Microsoft Foundry and public playgrounds; voice also on OpenRouter; LiveKit announced
What this means for you
For voice-agent teams, low-latency partial transcription and fast synthesis can reduce turn-taking delay without assembling separate providers. Published pricing and accessible APIs make an end-to-end latency and multilingual quality test practical.
Evidence & related material
Technical details & scores
| Transcription | 60 languages; continuous language detection; first partials just over 100 ms |
|---|---|
| Transcription Price | $0.54 per audio hour introductory price through year-end |
| Voice | 23 languages and 26 locales |
| Voice 2 1 Price | $22 per million characters |
| Voice Flash | 45 seconds of audio in 150 ms; $15 per million characters |
| Availability | Microsoft Foundry and public playgrounds; voice also on OpenRouter; LiveKit announced |
Technical impact 3
Engineering value 4
Adoption signal 3
Maturity 3
Verification cost 2
Risk 3