Microsoft releases MAI-Transcribe-1, the most accurate transcription model in the world

Microsoft has expanded its proprietary AI portfolio with the launch of MAI-Transcribe-1, a new speech-to-text model boasting a 3.9% Word Error Rate.

Following the successful launch of two in-house developed models, MAI-Voice-1 and MAI-Image-2, Microsoft today announced its third in-house developed model, MAI-Transcribe-1. Microsoft claims that MAI-Transcribe-1 is the most accurate transcription model in the world across 25 languages, with an average Word Error Rate of just 3.9%.

Out of the top 25 global languages (English, French, German, Italian, Spanish, Hindi, Portuguese, Czech, Danish, Finnish, Hungarian, Dutch, Polish, Romanian, Swedish, Japanese, Korean, Chinese, Arabic, Indonesian, Russian, Thai, Turkish, and Vietnamese), this new model from Microsoft ranks 1st by the FLEURS industry-standard benchmark in 11 core languages. It wins against Whisper-large-v3 on the remaining 14 and the recently released Google's Gemini 3.1 Flash on 11 of those 14.

The company claims that this model can maintain competitively high accuracy across all 25 supported languages, making it suitable for a wide variety of speech-to-text transcription use cases. It is important to note that this model doesn't support real‑time transcription, diarization, or biasing, but Microsoft plans to deliver these capabilities in an upcoming release.

Apart from the accuracy, this model is also very fast. Microsoft highlighted that the batch transcription speed of this new model is 2.5x that of the existing Microsoft Azure Fast offering. This model is now available in Microsoft Foundry starting at $0.36 USD per hour, which the company claims is the best price-performance of any large cloud provider. You can learn more about this new model here.

Today, the company also announced the availability of MAI-Image-2 and MAI-Voice-1 models in Microsoft Foundry. MAI-Voice-1 is priced at $22 USD per 1M characters, and developers can also create custom voices using this model through the Personal Voice feature in Azure Speech. The MAI-Image-2 image generation model is available for $5 USD per 1M tokens for text input and $33 USD per 1M tokens for image output.