MAI-Transcribe-2

MAI-Transcribe-2 is a speech recognition model from Microsoft AI that converts audio into accurate text across 60 languages with a single model. Diarization, word-level timestamps, contextual keyword biasing, and automatic language identification are built in rather than bolted on as separate services — and the model holds its accuracy in noisy environments and through code-switching between languages, Hinglish and Spanglish among them. It records an average word-error rate of 3.4% across the top 25 FLEURS languages and 5.2% across all 60, and transcribes an hour of audio in roughly ten seconds of model inference — 10× faster than GPT-Transcribe, 7× faster than Scribe v2, and 5× faster than Gemini 3.5 Transcribe.

Transcription has historically forced a choice between accuracy, latency, and cost — frontier accuracy meant slow batch pipelines and per-minute pricing that ruled out high-volume workloads entirely. MAI-Transcribe-2 collapses that tradeoff, defining the Pareto frontier for accuracy and latency on Artificial Analysis while priced at $0.10 per hour of audio through the end of 2026. It ships through Microsoft Foundry, the MAI Playground, and OpenRouter, joining MAI-Voice-2 and MAI-Image-2.6 in Microsoft AI’s in-house model family.