Microsoft AI model releases

Microsoft has expanded its AI capabilities through the release of several new models. The company first launched MAI-Transcribe-2, an AI transcription model designed for high-speed, multilingual speech-to-text tasks. Microsoft claims the model is up to 10 times faster than OpenAI’s GPT-Transcribe and five times faster than Google’s Gemini 3.5 Transcribe, featuring a 5.2% average Word Error Rate across 60 languages.

MAI-Transcribe-2 includes advanced capabilities such as speaker diarization to distinguish between different voices and word-level timestamps. Positioned as a cost-effective enterprise solution, the service has introductory pricing of approximately $0.10 per hour of audio, aimed at providing savings for high-volume users like call centers.

Following this, Microsoft released the MAI-Image-2.6 and MAI-Image-2.6-Flash models via its Foundry developer platform. These models support text-to-image generation and editing, with the Flash version optimized for high-volume, high-speed applications. While the models introduce new pricing structures for text and image tokens, reports have noted discrepancies in Microsoft’s official documentation regarding performance metrics and rankings in industry comparisons.