The Kaipher Editorial Team 
Microsoft has rolled out an automatic AI model selection feature in Visual Studio Code that will default to Anthropic’s Claude Sonnet 4 for GitHub Copilot coding tasks, marking a notable pivot from OpenAI’s GPT-5 despite a $13 billion investment in the latter since 2019 (PYMNTS). The update, deployed on September 15, will choose from Claude Sonnet 4, GPT-5, GPT-5 mini and other models; free users will rotate through available models, while paid subscribers will “primarily rely on Claude Sonnet 4,” per Microsoft.
Internal Benchmarks Drive Model Preference
According to an internal memo from Julia Liuson, president of Microsoft’s Developer Division, Claude Sonnet 4 outperformed GPT-5 in coding evaluations conducted earlier this year (The Verge). GitHub engineers reported that Claude “stays on track longer, understands problems more deeply, and provides more elegant code.” On the SWE-bench software engineering benchmark, Claude achieved 72–73% accuracy, edging out competing models. Microsoft sources reveal that developers were quietly directed to default to Claude for complex workflows months before the public announcement.
Expanding Beyond Visual Studio Code
The multi-model approach extends to Microsoft 365 Copilot, where early tests showed Claude delivering superior Excel automation and PowerPoint generation. Microsoft will source Anthropic models via Amazon Web Services—paying AWS for access despite its competition with Azure—to power Office applications without raising the $30 monthly Copilot fee (Axios).
Partnership Restructuring Amid AI Competition
On September 11, Microsoft and OpenAI signed a non-binding memorandum to restructure their partnership, enabling OpenAI’s transition to a public benefit corporation. Under the revised deal, OpenAI’s revenue share to Microsoft will drop from 20% to 8–10% by 2030, potentially adding over $50 billion to OpenAI’s coffers. Concurrently, Microsoft AI CEO Mustafa Suleyman unveiled plans to expand the company’s own infrastructure, with next-generation clusters “six to ten times larger” than the 15,000 Nvidia H100 chips used for the MAI-1 preview (TechBuzz).
Multi-Model Strategy Emerges
CEO Satya Nadella emphasized that Microsoft will “support a variety of models” across its products, citing Copilot’s auto-selection as a template for future offerings. The system optimizes model choice based on capacity and performance, offers a 10% request discount for paid users, and displays the selected model on hover. This strategic move underscores the intensifying competition in AI, where benchmark performance increasingly dictates partnerships, signaling a future of heterogeneous AI integrations within enterprise software.
