Microsoft Launches MAI-Transcribe-2: Faster, Second-Most Accurate, and Cheapest in Class

On September 3, Microsoft AI released the speech-to-text model MAI-Transcribe-2, announced by Microsoft AI chief Mustafa Suleyman and simultaneously listed on Artificial Analysis's non-streaming ASR leaderboard; it is available via Microsoft Foundry. The model lands in the market's top tier on accuracy, speed, and price, with the lowest pricing in the entire market.

Confirmed

• Accuracy: overall AA-WER of 2.0%, ranking second on the Artificial Analysis leaderboard, behind only Alibaba's Fun-Realtime-ASR-preview.
• Speed: up to 411x real time; per Suleyman, 10x faster than GPT-Transcribe and 5x faster than Gemini 3.5.
• Pricing: offered through Microsoft Foundry at $0.10 per hour, roughly $1.67 per thousand minutes of audio — lower than competitors like ElevenLabs and the lowest price on the market.

Why it matters

• Competition in the speech transcription market is heating up. Microsoft is entering with a combination of near-top accuracy, extremely fast inference, and aggressively low pricing, putting direct price and speed pressure on rivals like GPT-Transcribe, Gemini, and Alibaba's Fun-Realtime-ASR.
• The $1.67/thousand-minute pricing sharply lowers the barrier to large-scale audio transcription (podcasts, meetings, video captions) and could accelerate a price war in the industry.

2026-09-03 ~ 2026-09-03 · 7 related posts

• Microsoft's MAI-Transcribe-2 hits 2.0% WER at 411x real time for $1.67 per 1,000 minutes — ArtificialAnlys · 2026-09-03
• [source] Microsoft's MAI-Transcribe-2 hits ASR leaderboard at $0.10/hour, far below rivals — ArtificialAnlys · 2026-09-03
• [source] Microsoft's MAI-Transcribe-2 hits 2.0% WER at 411x real-time, priced at $1.67 per 1,000 minutes — ArtificialAnlys · 2026-09-03

4 near-duplicate retellings: jankulveit · mustafasuleyman · mustafasuleyman · altryne