Microsoft AI launches streaming transcription model and two voice models
Microsoft AI has launched MAI-Transcribe-2-Streaming, its first streaming transcription model, which the company says debuts at no. 1 on Artificial Analysis. It also announced two voice models, MAI-Voice-2.1 and a faster variant, MAI-Voice-2.1-Flash.
- Microsoft AI launched MAI-Transcribe-2-Streaming, its first streaming transcription model.
- According to Microsoft, the model debuts at no. 1 on Artificial Analysis.
- Two new voice models were announced: MAI-Voice-2.1 and MAI-Voice-2.1-Flash.
- Microsoft says the models are built for conversational voice agents and are accurate, fast and low cost.
- The announcement text gives no pricing, benchmark scores or availability details.
A first streaming transcription model
In its announcement on October 1, 2026, Microsoft AI introduced MAI-Transcribe-2-Streaming as its first streaming transcription model. According to the company, it debuts at no. 1 on Artificial Analysis, a ranking that Microsoft cites as evidence of its audio understanding capabilities.
Streaming transcription converts speech to text as it is spoken, instead of waiting for a complete recording. That makes it relevant for live, back-and-forth interactions such as voice agents, which is the use case Microsoft highlights for the release.
Two new voice models: MAI-Voice-2.1 and MAI-Voice-2.1-Flash
Alongside the transcription model, Microsoft AI announced MAI-Voice-2.1 and MAI-Voice-2.1-Flash. The company describes the second as its "blazing-fast variant" of the first. The page title also refers to multilingual voices, though the available text does not spell out which languages are supported.
Microsoft frames the three models as a set. In its words, they give users "the fast and fluid building blocks to create conversational experiences, with no compromise on accuracy or voice quality." The company positions the MAI-Transcribe and MAI-Voice models as accurate, fast and low cost, with chart-topping audio understanding and generation, aimed at building conversational voice agents.
What the announcement does not say
The available text of the announcement is brief. It does not include pricing, error rates, supported languages, latency figures or details on where and how developers can access the models. The claim of a top ranking comes from Microsoft itself, and the specific benchmark scores behind it are not given in the text provided.
What it means for builders of voice products
The release pairs speech recognition with speech generation, the two halves of a conversational voice agent. By offering a streaming transcription model together with a standard voice model and a faster Flash variant, Microsoft AI is presenting a matched set for teams that want to build spoken interfaces.
For developers, the practical questions will be access, cost and real-world latency, none of which the announcement text details. Until those are clear, the ranking on Artificial Analysis and Microsoft's own description of the models as accurate, fast and low cost are the main data points available.
