Microsoft Launches Proprietary Voice and Image AI Models…Also Cutting Prices Aggressively

Microsoft unveiled MAI-Transcribe-1, which recorded a 3.8% average word error rate across the top 25 languages on the FLEURS benchmark.

MAI-Voice-1 can generate 60 seconds of audio in under one second and supports personal voice cloning; its price is $22 per 1 million characters.

MAI-Image-2 landed among the top 3 on the arena.ai leaderboard, offers over 2× faster generation, and is priced at $5 per 1M tokens for text input and $33 per 1M tokens for image output.

WPP joined as an early corporate partner; Mustafa Suleyman said the voice model was built by about 10 people and the image team was under 10; Microsoft is applying aggressive pricing below Amazon and Google.

Microsoft's stock is down about 17% year-to-date; the company also said it will develop its own LLMs to pursue maximum efficiency and lowest cost when needed.