Microsoft AI releases MAI-Transcribe-2 speech model
Microsoft AI has released MAI-Transcribe-2, a 60-language speech recognition model that it says beats rival OpenAI, Google and ElevenLabs systems on speed and accuracy and costs $0.10 per hour at launch.
Microsoft AI announced MAI-Transcribe-2, the next version of its transcription model. The release adds speaker diarization, word-level timestamps, keyword biasing and configurable transcription styles.
The company says the model ranks first on the multilingual FLEURS benchmark across 60 languages, with an average word error rate of 5.2%. On Artificial Analysis, it sits on the Pareto frontier for accuracy and latency. It ranks second, not first, on that site's word-error-rate leaderboard, so Microsoft's "most accurate" claim depends on which benchmark you look at.
Microsoft says Artificial Analysis evaluations put the model's speed at 10x OpenAI's GPT-Transcribe, 7x ElevenLabs' Scribe v2 and 5x Gemini 3.5 Transcribe, with higher accuracy than all three. It says the speed advantage is largest on long-form audio.
The model offers two output styles. "Verbatim" keeps filler words and false starts for compliance and analysis work. "Clean" removes them for captions and published transcripts. Other features include:
- automatic language identification
- code-switching support for mixed-language speech such as Hinglish and Spanglish
- claimed robustness in noisy audio
Microsoft lists clinical note-taking, legal documentation, accessibility and closed captioning as target uses.
Launch pricing is $0.10 per hour of audio, offered for a limited time until the end of the year. Microsoft has not said what the price will be after that. The announcement also leaves out where developers can access the model, which languages are covered beyond the total count, and any results for real-time streaming.