On Tuesday, during the Build 2026 developer conference, Microsoft unveiled an innovative suite of proprietary AI models, accompanied by a plethora of additional updates.
These revelations, articulated during CEO Satya Nadella’s keynote address, encompass the entirety of the company’s product ecosystem, extending from silicon innovations to foundational software and cloud infrastructure.
In addition to the newly introduced AI models, notable highlights include Microsoft Scout, a cutting-edge personal agent tailored for workplace tasks, and the forthcoming Microsoft Surface Ultra laptop, engineered to efficiently handle substantial AI workloads locally.
The focal point of this new AI family is MAI-Thinking-1, Microsoft’s inaugural reasoning model. This mid-sized model, comprising 35 billion active parameters and boasting a 128K context window, is optimized for exceptional efficiency and performance, all while maintaining a modest token cost.
“MAI-Thinking-1 was meticulously crafted to excel in intricate multi-step instructions, extended context reasoning, and code generation,” remarked Kyle Daigle, the Chief Marketing Officer for Developers at Microsoft and Chief Operating Officer at GitHub, during a virtual media session preceding the keynote.
Daigle noted that MAI-Thinking-1 was constructed from the ground up using commercially licensed datasets.
He asserted that independent evaluators favored MAI-Thinking-1 over Anthropic’s Claude Sonnet 4.6 and that it equaled Claude Opus 4.6 on the SWE Bench Pro coding benchmark.
Included in the announcements were six additional MAI models that encompass capabilities in image generation, transcription, voice processing, and code:
How can you try the new MAI models?
According to Microsoft, MAI-Thinking-1 is currently accessible in Microsoft Foundry as a private preview.
The MAI-Image-2.5 models are already operational within PowerPoint and OneDrive, with an impending debut in Foundry. Moreover, MAI-Code 1 is available immediately in Copilot and VS Code.
MAI-Transcribe-1.5 is projected to launch shortly in 43 languages, while MAI-Voice-2 and its Flash variant are already available in 15 additional languages, featuring a variety of voice options.
What else was announced at Microsoft Build?
Microsoft also introduced Microsoft Scout, an anticipatory personal agent capable of managing scheduling, meeting preparations, and routine tasks via Teams and Outlook, all without requiring user prompts. This service begins its rollout to Frontier customers today.
In terms of hardware, the Surface RTX Spark Dev Box — equipped with NVIDIA’s RTX Spark chip — boasts the capacity to deliver up to one petaflop of AI computing power and comes with 128 gigabytes of unified memory, facilitating local operation of models with configurations up to 120 billion parameters. The device is slated for release later this year in the United States.
Additionally, Microsoft Discovery, the company’s platform for scientific research, is now generally available; Windows is being re-conceptualized as an agent-native runtime through a newly developed sandboxing system known as Microsoft Execution Containers, which is currently in preview.
Source link: Mashable.com.
