# Microsoft Launches MAI-Transcribe-1, Claiming World's Most Accurate AI Transcription Model

- Category: Flash (快讯)
- Published: 2026-04-07T11:52:00.000Z
- Source date: 2026-04-07
- HTML: https://www.88z.ai/en/news/flash-20260407-aa558936f3a95478
- Markdown: https://www.88z.ai/en/news/flash-20260407-aa558936f3a95478.md
- Other language: https://www.88z.ai/news/flash-20260407-aa558936f3a95478.md

Microsoft has released MAI-Transcribe-1, claiming it is the world's most accurate AI transcription model with a 3.9% word error rate. Supporting 25 languages, it outperforms Whisper-large-v3 and Gemini 3.1 Flash. Available on Microsoft Foundry, it offers batch transcription at 2.5 times the speed of Azure Fast services for $0.36 per hour, though real-time transcription and speaker separation are not yet supported.

## At a glance

- MAI-Transcribe-1 claims to be the world's most accurate model with a 3.9% word error rate, outperforming Whisper and Gemini.
- The model is live on Microsoft Foundry with fast batch processing but currently lacks real-time transcription and speaker separation.

## The update

Microsoft has officially launched the AI transcription model MAI-Transcribe-1. The company claims it achieved a 3.9% word error rate in FLEURS benchmarks, making it the most accurate transcription model globally. It supports 25 major languages and ranks first in performance evaluations for 11 core languages, outperforming well-known models like Whisper-large-v3 and Gemini 3.1 Flash. Regarding efficiency and cost, MAI-Transcribe-1 processes batch transcriptions at 2.5 times the speed of Azure Fast services, priced at $0.36 per hour. The model is now available via the Microsoft Foundry platform. Notably, the current version does not support real-time transcription or speaker separation, focusing primarily on batch processing scenarios.

## Sources

- [IT之家](https://www.ithome.com/0/935/617.htm) (2026-04-07)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.