Microsoft Launches MAI-Transcribe-2-Streaming, Its First Real-Time Streaming Speech-to-Text Model

Microsoft announced on October 1 the release of MAI-Transcribe-2-Streaming, its first real-time streaming speech-to-text model. The model outputs text continuously while a person is speaking and supports 60 languages with automatic language detection. Developers can access it through Microsoft Foundry, MAI Playground, and OpenRouter.

During the promotional period, the model is priced at $0.54 per hour, roughly $9 per 1,000 minutes. In Artificial Analysis benchmarks, it achieved a final transcription latency of 0.13 seconds and a word error rate of 2.50%, ranking first among 28 streaming speech-to-text models.

The model features continuous automatic language detection, so users do not need to specify a language beforehand. Microsoft said the model also enables applications to start understanding intent and trigger reasoning or tool calls before a speaker finishes a sentence.

Microsoft previously launched the non-streaming MAI-Transcribe-2 model in Azure Speech public preview at $0.10 per hour.