Microsoft unveils in-house AI models and agents at Build conference as it moves away from OpenAI partnership

Microsoft announced MAI-Thinking-1, its first reasoning model, along with six other new models focused on image, voice, transcription, and coding. The company stated the medium-size MAI-Thinking-1 is 'built from scratch for serious math, coding, and real-world enterprise deployment' and is cheaper than OpenAI equivalents on some tasks, developed without distillation (training using another company's AI model).

Microsoft is promoting a Copilot 'super app' that integrates AI agents called 'Autopilots,' designed to perform tasks like reviewing emails, joining Teams chats, checking calendars, and sending daily briefings. The first Autopilot offering, called 'Scout,' is billed as 'your always-on personal agent' and is marketed as 'autonomous, long-running agents with full enterprise compliance.'

AI chief Mustafa Suleyman stated the goal is to 'prove that we can become one of the top four labs in the world,' noting that 'the pivotal moment was renegotiating our contract with OpenAI' to train models at larger scale with Microsoft's own IP and data. The company emphasized its 'humanist superintelligence' approach and advantage of existing enterprise client relationships and Azure resources.