Microsoft unveiled its own reasoning model, AI agents and cybersecurity tools at its annual Build developer conference, shifting its AI strategy toward internally developed models.

At the event, Microsoft introduced its first reasoning model, MAI-Thinking-1, and 6 models for images, voice, transcription and coding.

The Verge reported on June 3 that Microsoft AI CEO Mustafa Suleyman (무스타파 술레이만) said, "The goal is to prove that Microsoft can be one of the world’s top four research labs."

He named Google DeepMind, OpenAI and Anthropic as major AI research companies, and said Microsoft is not yet among them.

Microsoft explained that MAI-Thinking-1 is a model built from scratch targeting mathematics, coding and real enterprise deployments. It said the model was trained on Microsoft’s own data and intellectual property without distillation that uses other companies’ models. It also said the model is cheaper than comparable OpenAI models for some tasks.

Suleyman stressed that after revising its contract with OpenAI, Microsoft can pursue training larger models and developing independent superintelligence using its own intellectual property.

Microsoft put security and agents at the forefront of its push into the enterprise market. CEO Satya Nadella introduced a recently launched AI cybersecurity tool, MDASH, and said bundling 100 AI agents finds vulnerabilities better than a single model.

It is unclear how much success Microsoft will achieve with its in-house models. Microsoft highlighted benchmark results for the 7 new models, but The Verge said there is no guarantee those figures will lead to real adoption.