Microsoft used its Build 2026 developer conference on Tuesday to announce a new family of in-house AI models, alongside a slew of other news. The announcements, delivered during CEO Satya Nadella’s conference keynote, span the company’s full product stack, from silicon to operating system to cloud infrastructure. Besides the new AI models, highlights include Microsoft Scout, a new personal agent for workplace tasks, and an upcoming Microsoft Surface Ultra laptop designed to run large AI workloads locally.
The centerpiece of this new family is MAI-Thinking-1, Microsoft’s first reasoning model. It’s a mid-sized, 35 billion active parameter model with a 128K context window built for high efficiency and performance, but importantly, at a low-token cost. “MAI-Thinking-1 was designed to be good at complex multi-step instructions, long context reasoning, and code generation,” said Kyle Daigle, Microsoft Developer CMO and COO of GitHub, at a virtual media briefing ahead of the keynote.
According to Daigle, MAI-Thinking-1 was built from scratch on commercially licensed data. The company says independent evaluators preferred it over Anthropic’s Claude Sonnet 4.6, and that it matches Claude Opus 4.6 on the SWE Bench Pro coding benchmark. Six additional MAI models were announced, covering image generation, transcription, voice, and code:
The new models are:
How can you try the new MAI models?
Per Microsoft, MAI-Thinking-1 is available in Microsoft Foundry as a private preview. The MAI-Image-2.5 models are already live in PowerPoint and OneDrive, and will be arriving soon in Foundry. MAI-Code 1 is available now in Copilot and VS Code.
Mashable Light Speed
Microsoft said that MAI-Transcribe-1.5 will be available soon in 43 languages, while MAI-Voice-2 and a Flash variant are already available in 15 additional languages with multiple voice options.
Eventually, all of the models will be available in Foundry and a new dedicated environment, dubbed MAI Playground.
What else was announced at Microsoft Build?
Also announced was Microsoft Scout, a proactive personal agent that handles scheduling, meeting prep, and routine tasks through Teams and Outlook without waiting for user input. It begins rolling out to Frontier customers today.
On hardware, the Surface RTX Spark Dev Box — powered by NVIDIA’s RTX Spark chip — delivers up to one petaflop of AI compute and 128 gigabytes of unified memory, with the stated ability to run models up to 120 billion parameters locally. It ships later this year in the US.
Rounding out the announcements: Microsoft Discovery, the company’s scientific research platform, is now generally available; and Windows is being repositioned as an agent-native runtime through a new sandboxing system called Microsoft Execution Containers, now in preview.
