Microsoft has unveiled seven new artificial intelligence models, marking one of its biggest efforts yet to build foundational AI technology internally even as it remains one of OpenAI’s closest partners and largest investors.

Announced at the company’s Build developer conference, the new models span reasoning, coding, image generation, voice and transcription. “The compute used to train frontier models has increased by a factor of one trillion. Now we expect another thousand-fold increase over the next three years, which in turn means more advanced capabilities, and the continued rollout of ever more effective AI,” the company noted.

At the centre of the announcement is MAI-Thinking-1, Microsoft’s first reasoning model. The company said the model is designed for tasks that require multiple steps of analysis and problem-solving, particularly in software engineering and technical workflows. According to Microsoft, the model performs competitively on coding benchmarks and was trained using commercially licensed data. “It matches leading models on key software engineering benchmarks, and demonstrates advanced mathematical reasoning capabilities, and is preferred to (Claude’s) Sonnet 4.6 in our blind human side-by-side evaluations. We trained it from the ground up on clean data, without distillation from third-party models.”

Microsoft also introduced MAI-Code-1-Flash, a smaller coding-focused model aimed at helping developers write, review and edit software more efficiently. The company plans to integrate the model into developer tools including GitHub Copilot and Visual Studio Code.

The new portfolio extends beyond coding. Microsoft launched MAI-Image-2.5 and MAI-Image-2.5-Flash, two models focused on image creation and editing, as well as MAI-Voice-2 and MAI-Voice-2-Flash, which expand the company’s speech-generation capabilities across multiple languages. Completing the lineup is MAI-Transcribe-1.5, a speech-to-text model designed to improve transcription speed and accuracy.

Independent from OpenAI partnership

The releases highlight Microsoft’s growing ambitions as a model developer in its own right. Since the generative AI boom began, the company has largely been viewed as the primary commercial partner of OpenAI, embedding its technology across products ranging from Copilot to Azure. At the same time, Microsoft has quietly expanded its own AI research efforts and explored ways to build alternative model families for different use cases.

That strategy mirrors a broader shift across the industry. As demand for AI services grows, companies are increasingly looking to develop proprietary models that offer greater control over costs, deployment and product roadmaps.

Mustafa Suleyman, chief executive of Microsoft AI, said the company’s focus is not solely on releasing larger or more capable models, but on building systems that continuously improve through training, evaluation and reinforcement learning. “All these models are built on a shared foundation, hill-climbing from the bottom with zero distillation. They share the same data discipline, the same infrastructure and the same evaluation framework,” he said.

The new MAI family will be available across Microsoft’s AI ecosystem, including Azure AI Foundry and Copilot products.