Microsoft Unveils Three New In-House AI Models

Microsoft unveiled three new in-house artificial intelligence models on Thursday, marking a significant step in its efforts to develop proprietary technology for transcription, voice generation, and image creation. The models, named MAITranscribe1, MAIVoice1, and MAIImage2, are now available through the Microsoft Foundry platform and the MAI Playground, allowing developers and businesses to test, customize, and deploy them across various applications.

MAITranscribe1 focuses on speech-to-text transcription and supports 25 languages, including English, French, German, Spanish, Hindi, Japanese, Korean, Chinese, Arabic, and Russian. Microsoft states that the model achieves an average word error rate of 3.9 percent, outperforming competitors such as OpenAI's Whisper large v3 across all supported languages and Google's Gemini 3.1 Flash in 22 of 25 languages, according to the FLEURS industry benchmark. It processes audio files up to 200 megabytes in formats like MP3, WAV, and FLAC, with batch transcription speeds 2.5 times faster than Microsoft's previous Azure Fast offering. Priced at $0.36 per hour, the model is designed for uses such as generating instant transcripts for Microsoft...