We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results coming soon -- stay tuned! Lots of details in our technical blog post. microsoft.ai/news/frontier-p…

Great to see Microsoft-Decision-1 as the smartest model from the most attractive quadrant, with the highest non-hallucination rate among decision models. A strong fit for real customer use cases, which is what we trained it for.

Today we are announcing our biggest bet yet on our own silicon. Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services. And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems. We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot. Our goal is to build on the full-stack efficiency gains we’ve already seen, while continuing to bring the latest innovations from across the ecosystem into our fleet.

Starting today, the full power of Microsoft Office is rolling out to the Copilot App in Frontier. We're bringing together the capabilities that your business runs on every day within Word, Excel and PowerPoint inside of the new OS for work. This new experience supports your entire workflow, from the first prompt to the finishing touches. You can collaborate with others, use branded templates, and return to earlier iterations through version history as your work evolves. In the same workspace, you can choose from a robust library of frontier models or use the Auto router to match the right model to the task at hand. Learn more in my blog post: techcommunity.microsoft.com/…

Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality. We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.

This is your reminder that there are tens of millions of people whose primary experience with AI is Microsoft Copilot We are still so early

You may not like to hear this, but Microsoft is going to ram Copilot — the new “operating system for work” as @satyanadella called it today — down the throat of every enterprise in the world. It’s going to be a damn good harness. This strategy is going to work and they will be a $5t company within a year. They’ve been quietly adding brilliant product leaders (notably @jacobandreou) who has better design instincts than almost anyone else in large cap tech today. He was Evan’s key deputy at Snap and responsible for almost everything that worked on their march through hell to 500m DAUs. Microsoft finally started to ship decent models this year. MAI is not SOTA, but it’s a class of models that are efficient, specialized, and tunable. For complex work, they send you off to ChatGPT or Claude, which they will happily serve you on Azure, a business that is ripping with 43% yoy growth. Muse and (and now Grok Bot) have shown us that a great harness doesn’t require an extraordinary model to be game-changing. Ecosystem integration, disciplined product design, access to tools, and more than anything else — user trust — is what it takes. Anyway, I think about this chart a lot.