Microsoft is capitalizing on legal fears to sell its powerful new AI model to businesses

Microsoft is abandoning a family of seven new AI models, the company announced Tuesday during the opening keynote at Build, its annual developer conference.

The highlight of the blockbuster is the 35-billion-parameter MAI-Thinking-1, which Microsoft AI leader Mustafa Suleiman described on stage during the keynote as Microsoft’s « first reasoning model » and said independent early testers « preferred it as an overall quality, side-by-side, against (Anthropic’s Claude) Sonnet 4.6. » MAI-Thinking-1 also scored 97% on the AIME benchmark, which measures advanced mathematical and problem-solving abilities, and « most important of all, » 53% on the SWE Bench Pro, which measures AI agents’ ability to handle complex coding tasks. Anthropic’s Claude Opus 4.6 currently has 51.9%, but OpenAI’s GPT-5.4 scored 59.1%, according to data by Scale Labs, the model performance tracking division of Scale AI.

The big selling point was that MAI-Image-1 was trained « entirely from the ground up, » as Suleiman put it. That is, it is trained solely to be an exceptional model of reasoning in a set of possible tasks, rather than to perform well on specific benchmark tests. It’s also trained « with absolutely zero distillation, » according to Suleiman. « Distillation » is AI industry jargon for using another model, perhaps created by a competing company, to train a new one; in this way, models can become a sort of scaffolding for other developers to use and build upon.

Microsoft’s pitch to enterprise customers seems to be basically that distillation could lead to problems in the future because there is so much uncertainty and legal wrangling around the issue of sourcing training data and potential copyright infringement. The more clearly you can trace the origin of data within the AI ​​tools you use, the better. MAI-Image-1 « is built with enterprise-grade, clean and commercially licensed data, which means you can put it into production in a highly reliable manner with complete confidence. » We’ll have to wait and see for a more detailed explanation of exactly how Microsoft has licensed all the training data for this model, but that’s the gist of the business.

The other models Suleiman introduced during the Build keynote included MAI-Image-2.5 and MAI-Image-2.5 Flash, image generation models that « provide a step change in quality » that, at the time of writing, is in place number three on the text-to-image scoreboard of the Arena.AI model, just behind Google’s Nano Banana 2; MAI-Transcribe-1.5, which Suleiman said is « the best transcription model in the world »; speech generation models MAI-Voice-2 and Mai-Voice-2 Flash; and MAI-Code-1-Flash for code generation.

Taken together, this is the biggest AI product announcement from Microsoft since it debuted the first fully in-house models, MAI-Voice-1 and MAI-Voice-1-preview, in August. The company’s AI efforts have relied heavily on its partnership with OpenAI before, but the two companies have since gradually drifted apart.

However, its early investments in OpenAI helped Microsoft gain some traction in the early days of the AI ​​boom; it is one of the few old-school tech giants that has successfully managed to become a leader in this race, while others like Apple and IBM lag behind. The company brands its AI ambitions under the motto « humanistic superintelligence, » through which it tries to position itself as a technology company that always puts human interests and agency first: « We want to both explore and prioritize how the most advanced forms of AI can keep humanity in check while accelerating our way to addressing our most pressing global challenges, » Suleiman wrote in a company blog post in November.