Nadella says Microsoft’s MAI models now outperform frontier AI at lower cost

Microsoft ↗ 0.03% CEO Satya Nadella laid out what he called a "Frontier Diffusion and Control" strategy in a blog post on Wednesday, detailing how the company's proprietary MAI family of AI models is now being deployed across GitHub Copilot, Excel, Outlook, and other Microsoft 365 applications to reduce inference costs while matching or exceeding the performance of models from OpenAI and Anthropic.indianexpress+2

The approach routes high-volume, routine tasks to smaller, task-optimized MAI models while reserving expensive frontier models for the most demanding workloads — a multi-model orchestration system designed to lower the cost of running AI at scale across Microsoft's product suite.lumienai+1

Early Results in Copilot and Excel

Microsoft shared production data showing that MAI-Code-1-Flash, a lightweight coding model with roughly 5 billion active parameters, achieved an approximately 10 percent higher code acceptance rate than GPT-5.4 Mini and Claude Haiku 4.5 in Visual Studio Code, while consuming 10 percent fewer median tokens. Users of the model were also 6 percent more likely to return across multiple days compared to GPT-5.4 Mini, a figure that rose to 11 percent against Claude Haiku 4.5.remio+1

In Excel, Microsoft adapted a MAI checkpoint using reinforcement learning environments tailored to spreadsheet workflows. The company said the resulting model performs on par with GPT-5.6 on common Excel tasks at lower operating cost, though it emphasized that the comparison applies to frequently occurring workflows rather than every reasoning challenge.firstpost+1

"We are now seeing MAI models outperform general-purpose frontier models in many use cases while using a fraction of the tokens," Nadella wrote.indianexpress

A Broader Shift Away from Third-Party Dependence

The announcement extends a trajectory Microsoft set in motion at Build 2026 on June 2, when its AI division unveiled a family of seven MAI models spanning reasoning, coding, image generation, voice, and transcription — all trained from scratch without distillation from third-party providers. Bloomberg reported earlier in July that Microsoft had already begun replacing some OpenAI and Anthropic model usage in Excel and Outlook, with tens of thousands of weekly prompts rerouted to MAI models.linkedin+4

During a prior presentation, Microsoft AI CEO Mustafa Suleyman said the company wanted to reduce and ultimately eliminate spending on Anthropic models. Microsoft has stressed, however, that it is not abandoning external providers. Nadella described the strategy as complementary, with OpenAI and Anthropic models remaining available through Azure AI Foundry for tasks where they outperform in-house alternatives.pureai+2

Microsoft said it plans to extend the same reinforcement-learning-based optimization approach to Copilot Chat, PowerPoint, and additional agentic products in the coming months, while making the framework available to enterprise customers through Microsoft Foundry.indianexpress+2