Microsoft AI’s MAI-Transcribe-2 undercuts OpenAI, Google and ElevenLabs on price and speed

Microsoft AI unveiled MAI-Transcribe-2, a speech-recognition model claiming to outperform OpenAI, Google, and ElevenLabs in terms of speed, accuracy, and cost. Priced at $0.10 per hour of audio, the model represents a significant 72% price reduction from its initial $0.36 an hour offering. This cost-cutting for large-scale transcription at enterprise volumes, such as a bank or telecom processing 100,000 hours annually, transforms transcription from a negotiable line item to a predictable expense.

MAI-Transcribe-2 supports transcription in 60 languages, up from 43 and 25 in previous models, and is designed for real-world, noisy audio conditions, making it more applicable to business scenarios than clean studio recordings. Key features include speaker diarization, which identifies who spoke in a multi-person recording, word-level timestamps for precise search and video alignment, keyword biasing for specialized terminology, and automatic language identification.

The model's standout features include a configurable output style with verbatim and clean modes for compliance needs, and code switching capabilities to handle conversations that switch languages mid-sentence. This comprehensive set of features is offered at a single price point, a stark contrast to specialty vendors charging premiums for each capability separately.

Microsoft benchmarks MAI-Transcribe-2 on three fronts: FLEURS, Artificial Analysis, and raw speed. While the model's average word error rate of 5.2% on FLEURS might seem low, it has increased from 3.7% in June due to broader language coverage. On Artificial Analysis, the model has climbed to second place, directly challenging ElevenLabs' Scribe v2 as the top performer.

Speed-wise, the model is claimed to be 10 times faster than OpenAI's GPT-Transcribe, seven times faster than ElevenLabs Scribe v2, and five times faster than Google's Gemini 3.5 Transcribe, all of which are supported by independent evaluations.