Microsoft Ditches OpenAI for Cheaper In-House AI

Microsoft is now using its own AI models in products like GitHub Copilot, Excel, and Outlook. This shifts away from models built by OpenAI and Anthropic. Here's the thing: CEO Satya Nadella explained this move is primarily about cost. Microsoft's in-house MAI models can handle many tasks more efficiently and cheaply than larger, general-purpose models. For example, its tuned model for Excel reportedly matches GPT 5.4 but runs up to ten times more efficiently. MAI-Code-1-Flash, designed for GitHub Copilot, is also said to be comparable to other models but cheaper. What's interesting is that these MAI models are trained within the actual product environment, learning from real customer tasks. This approach helps ensure they perform well where it matters most. The bottom line: Microsoft aims to use expensive "frontier" models only for the most complex problems, routing other tasks to its own more cost-effective AI. This could lead to more efficient and affordable AI integration across many software platforms.