Microsoft Launches MAI-Transcribe-2 With 10x Faster Speech Recognition
Microsoft has announced MAI-Transcribe-2, its latest speech recognition model. The company claims it is its fastest, most accurate, and cheapest transcription model yet.
MAI-Transcribe-2 improves accuracy across 60 languages
MAI-Transcribe-2 ranks first on the FLEURS multilingual benchmark across 60 languages, achieving an average Word Error Rate of 5.2%.
Microsoft also says the model ranks second on the independent Artificial Analysis WER leaderboard.
According to the company, MAI-Transcribe-2 delivers higher transcription accuracy than competing models from OpenAI, Google, and ElevenLabs.
Microsoft claims MAI-Transcribe-2 is up to 10x faster
Speed is another major focus of the new model.
Microsoft claims MAI-Transcribe-2 is up to 10 times faster than OpenAI’s GPT-Transcribe. It is also reportedly seven times faster than ElevenLabs Scribe v2 and five times faster than Google’s Gemini 3.5 Transcribe.
These performance improvements could make the model particularly useful for applications that need to process large amounts of audio or generate transcripts quickly.
New transcription features include speaker diarization and code switching
MAI-Transcribe-2 includes speaker diarization, allowing it to distinguish between different speakers in a conversation.
The model also supports word-level timestamps, keyword biasing, and automatic language identification.
Microsoft has added support for code-switching as well, allowing MAI-Transcribe-2 to handle conversations that move between multiple languages. The company specifically highlights examples such as Hinglish and Spanglish.
Verbatim and Clean modes offer different transcription styles
MAI-Transcribe-2 comes with two transcription modes depending on how users want the output formatted.
Verbatim mode preserves filler words, false starts, and other parts of natural speech.
Clean mode removes those elements to produce more readable captions, meeting notes, and transcripts.
MAI-Transcribe-2 costs $0.10 per hour
Microsoft is offering MAI-Transcribe-2 at a promotional price of $0.10 per hour of audio.
The promotional pricing will remain available until the end of 2026. Microsoft has not yet announced how much the model will cost starting in 2027.
MAI-Transcribe-2 is available now through Microsoft Foundry and the MAI Playground. The release follows Microsoft’s recent launch of MAI-Code-1.1-Flash.
Competition across the AI model market also continues to accelerate. GPT-6 Astra has launched, while Google recently introduced Gemini 3.8 Flash.
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
User forum
0 messages