Microsoft prefers Anthropic's AI model for its coding tool
What's the story
Microsoft has updated its Visual Studio Code editor with an AI model selector that will automatically pick the best model for "optimal performance." This new facility will select between Claude Sonnet 4, GPT-5, GPT-5 mini, and other models for free GitHub Copilot users. However, paid users will "primarily rely on Claude Sonnet 4."
Strategic shift
Microsoft's internal benchmarks favor Claude Sonnet 4
Julia Liuson, head of Microsoft's developer division, had said in an internal email in June that "based on internal benchmarks, Claude Sonnet 4 is our recommended model for GitHub Copilot." This was before the launch of GPT-5 but indicates a strategic shift toward Anthropic's AI models. Sources familiar with Microsoft's developer plans have also revealed that the company has been encouraging its developers to use Claude Sonnet 4 in recent months.
AI development
Microsoft is training its own AI models
Along with the shift toward Anthropic's AI models, Microsoft is also investing heavily in training its own. Mustafa Suleyman, Microsoft's AI chief, revealed at an employee-only town hall last week that "we're also going to be making significant investments in our own cluster." He noted that MAI-1-preview was only trained on 15,000 H100s—a small cluster considering the bigger picture.
Future plans
Integration of Anthropic's models into Microsoft 365 apps
Microsoft is also said to be planning the integration of Anthropic's AI models into some features of its Microsoft 365 apps. The Information reports that the Microsoft 365 Copilot will be "partly powered by Anthropic models," after tests showed such models outperformed OpenAI in Excel and PowerPoint. This further highlights Microsoft's growing preference for Anthropic over OpenAI in its software development and productivity tools.