Editor's take: Microsoft has long been the financial lifeline of OpenAI, but its growing reliance on Anthropic's models suggests that loyalty may be giving way to performance. By favoring Anthropic in key products, Microsoft shows that funding OpenAI doesn't guarantee exclusivity – a reminder that Redmond's AI strategy is ultimately pragmatic, not sentimental.
Microsoft is signaling a shift in its AI priorities, favoring Anthropic's Claude Sonnet 4 over OpenAI's GPT-5 models in its flagship developer tool, Visual Studio Code. The company is introducing an automatic AI model selection feature for GitHub Copilot that chooses the optimal model for coding assistance. Paid users will primarily rely on Claude Sonnet 4, while free users will see a mix, including GPT-5 and GPT-5 mini.
Sources familiar with Microsoft's developer plans told The Verge the company has directed engineers to use Claude Sonnet 4 for months. In June, Microsoft's developer division head, Julia Liuson, said that the company prefers Anthropic's AI model for its GitHub tool because of its performance.
"Based on internal benchmarks, Claude Sonnet 4 is our recommended model for GitHub Copilot," she said.
Although her statement came months before GPT-5's August release, Microsoft's model guidance has not changed.
The move marks a significant moment for Microsoft, which has invested heavily in OpenAI over several years. The company has poured more than $13 billion into OpenAI since 2019 and maintains a complex revenue-sharing arrangement with the startup, creating a complex mix of competing interests.
Last week, the two companies announced a deal that could pave the way for OpenAI's initial public offering. The agreement allows the AI startup to leverage rival cloud providers, and Microsoft expects to provide further details on the next phase of the partnership soon.
Despite this, Redmond is hedging its bets. Mustafa Suleyman, Microsoft's AI chief, revealed at a recent employee town hall that the company is making significant investments in its own AI infrastructure. He noted that the company trained the MAI-1-preview model exclusively on 15,000 H100 GPUs, a modest cluster by industry standards.
Microsoft's partial pivot toward Anthropic also extends beyond coding. Some Microsoft 365 apps, including Excel and PowerPoint, will soon partly rely on Anthropic models after internal testing suggested these models outperformed OpenAI alternatives in productivity scenarios.
The decision highlights Microsoft's evolving AI strategy, balancing its long-standing OpenAI partnership with alternatives that deliver measurable performance gains. For developers, this could mean a more streamlined experience in Visual Studio Code and improved productivity in Microsoft 365 apps.
Automatic model selection signals Microsoft's focus on practical performance over corporate allegiance. This change represents a subtle but crucial realignment in the AI tools that power its software ecosystem and reflects the company's broader strategy for combining internal and external AI models.