Microsoft Launches New AI Models Amid Complicated Partnership with OpenAI
Microsoft has unveiled two new in-house AI models, MAI-Voice-1 AI and MAI-1-preview, setting the stage for competition against established models like GPT-5 and DeepSeek, reports 24brussels.
The newly released MAI-Voice-1 model excels in generating audio, producing a minute’s worth of sound in under one second using a single GPU. The MAI-1-preview model offers insights into future features for the Copilot assistant, enhancing user experience.
These innovations are already integrated into existing Microsoft applications, powering features such as Copilot Daily, which utilizes AI to recite daily news, and generating discussions to elucidate various topics. Users can test MAI-Voice-1 in Copilot Labs, adjusting the voice and delivery style of the AI model.
MAI-1-preview, trained on approximately 15,000 Nvidia H100 GPUs, is tailored for users who require an AI capable of following instructions and providing helpful responses to routine queries. Microsoft AI chief Mustafa Suleyman emphasized that the company aims to prioritize consumer-driven applications, stating, “So, we have vast amounts of very predictive and very useful data on the ad side, on consumer telemetry, and so on.”
Looking ahead, Microsoft plans to incorporate MAI-1-preview into its Copilot AI, currently utilizing OpenAI’s language models. The model is also undergoing public testing on LMArena, a platform for AI benchmarking.
In a recent blog post, Microsoft expressed its ambitions, stating, “Not only will we pursue further advances here, but we believe that orchestrating a range of specialized models serving different user intents and use cases will unlock immense value.” The introduction of these AI models reflects Microsoft’s strategic pivot towards competing more directly in the AI landscape.
