According to a recent Bloomberg report, Microsoft has quietly built a small army of domain-specific models, detailed at its Build developer conference in June, which are now slowly but surely replacing OpenAI's models as the power behind the AI features in Microsoft products. Microsoft describes MAI-Thinking-1 as a “medium-sized model that stands among the strongest models in its weight class” and says it "matches leading models on key software engineering benchmarks, demonstrates advanced mathematical reasoning capabilities, and is preferred to Sonnet 4.6 in our blind human side-by-side evaluations.” The shift reflects that while OpenAI and Anthropic build ever-larger models capable of tackling almost any task, it is cheaper and easier to train a small domain-specific model that can run dozens of instances on a single accelerator. Microsoft can now replace a frontier model with more precise tools that do the same job as quickly and cheaply as possible, as bean-and-token-counters aren't yet sure if it's possible to sell AI at a profit. Smaller models free up memory and improve hardware utilization, allowing Microsoft to deploy the right AI for the right job at the right time, such as spinning up more instances of the best model for speech-to-text traffic while keeping costs controlled.