Building a new class of safer, more capable AI systems we call Humanist Superintelligence: AI that is always aligned, controllable, and in service of humanity.

Joined July 2017
Photos and videos
Pinned Tweet
Seven new models launching at Build: let’s go! Reasoning. Code. Image. Transcribe. Voice. Built from scratch on a clean data lineage, designed for efficiency, working seamlessly as a family of models Thread 🧵 #MSBuild
137
524
3,366
388,527
What does it take to build coding models that meet developers where they work? Go behind the scenes with Microsoft AI to explore how we build and optimize code. From training and evaluation to performance, safety, and real-world developer feedback. msft.it/6012viG1u
6
7
62
7,642
MAI-Code-1-Flash is now rolled out to 100% of GitHub Copilot Free, Education, Pro, Pro , and Max subscribers in VS Code. Copilot CLI roll out and Enterprise/Business preview on its way. Give it a try and let us know what you think!
37
95
797
124,741
Let’s go a bit deeper into Frontier Tuning launched at Build and see a live demo! Frontier Tuning is how we enable you to develop custom AI by building a reinforcement learning environment (RLE) to hill climb on your data, tools, and knowledge. Think of it like a training gym for your AI.
10
25
226
11,140
Microsoft AI retweeted
I have been crushing through some code including advanced research, planning, and implementation with MAI-Code-1-Flash and I am loving it. Token usage is efficient and cost is fantastic. Flipping bugs for around 20 to 30 AIC each.
15
8
112
17,972
Fun fact – the images announcing MAI-Image-2.5 - our image generation and editing model were generated by the model itself! Strong photo-realism, precise edits, quality down to the fine details. Try it on our Playground: msft.it/6013vdFuH or on @OpenRouter and share your favorite images on this thread
24
26
235
13,900
Seven new models launching at Build: let’s go! Reasoning. Code. Image. Transcribe. Voice. Built from scratch on a clean data lineage, designed for efficiency, working seamlessly as a family of models Thread 🧵
189
185
4,832
28,753,169
MAI-Voice-2 MAI-Voice-2-Flash. Our most expressive text-to-speech model yet: - 15 languages with emotion control (excited, whispered, embarrassed and more) - Stable speaker identity across long-form content - Code-switching for Hindi-English and Spanish-English - Preferred over MAI-Voice-1 in 72% of evals MAI-Voice-2-Flash coming soon, built for real-time agents and high-volume workloads where latency and cost matter most.
2
1
18
5,612
Seven models. One lab. Zero distillation. Try them on Microsoft Foundry. Hold onto your hats, let’s go!
6
2
32
5,026