By adding an automatic AI model selection option that favors Anthropic’s Claude Sonnet 4 over OpenAI’s GPT-5 for coding jobs, Microsoft has made a dramatic strategic shift in Visual Studio Code. Despite the tech giant’s $13 billion investment in OpenAI since 2019, this development points to a more significant change in its AI strategy.
Users of GitHub Copilot will be able to pick between Claude Sonnet 4, GPT-5, GPT-5 small, and other models automatically thanks to the new auto-model selection option, which was introduced on September 15. According to Microsoft’s official release, paying members would “primarily rely on Claude Sonnet 4,” while free-tier users will see a rotation of models.
Model Preference Is Determined By Internal Benchmarks
Internal performance reviews that have continuously preferred Anthropic’s offering are the reason for the decision. In a June internal email, Julia Liuson, the head of Microsoft’s developer division, informed staff that “Claude Sonnet 4 is our recommended model for GitHub Copilot based on internal benchmarks.” This suggestion was made before GPT-5 was released, and it is said that it hasn’t altered even since OpenAI released their most recent model.
According to Microsoft sources with knowledge of the company’s development strategy, programmers have been subtly told for months to use Claude Sonnet 4 by default for coding tasks. Claude 4 “stays on track longer, understands problems more deeply, and provides more elegant code,” according to GitHub workers who conducted early testing on the platform.
One important element was the model’s exceptional performance on challenging coding benchmarks. On the software engineering benchmark SWE-bench, Claude Sonnet 4 outperformed rival models with an accuracy of 72–73%. Anthropic’s model surpassed OpenAI’s solutions in “subtle but important ways” for particular development workflows, according to Microsoft’s internal tests.
Going Beyond Visual Studio Code
This variety of AI models is included throughout Microsoft’s entire productivity suite, not just Visual Studio Code. The Information stated that Claude will soon be “partly powered by Anthropic models” in Microsoft 365 Copilot after internal testing revealed he was more proficient in PowerPoint presentation creation and Excel automation.
Even though Amazon Web Services is a direct rival of Microsoft’s Azure cloud platform, Microsoft will pay AWS to use Anthropic’s models in a convoluted deal that highlights the competitive realities of the AI industry. Microsoft intends to keep Copilot’s $30 monthly pricing for users in spite of this extra expense.
Restructuring Partnerships In The Face Of AI Competition
The time aligns with major shifts in Microsoft and OpenAI’s partnership. In order to restructure their relationship and enable OpenAI to pursue its conversion to a public benefit company, the two businesses signed a non-binding memorandum of understanding on September 11. By 2030, OpenAI’s revenue share with Microsoft is anticipated to decrease from 20% to roughly 8–10% under the updated conditions, potentially generating over $50 billion more in income for the AI startup.
At a town hall meeting on September 12, Microsoft AI CEO Mustafa Suleyman revealed intentions for “significant investments” in the company’s own AI infrastructure. Suleyman described the 15,000 Nvidia H100 chips used to train the company’s MAI-1-preview model as “tiny” in comparison to the “six to ten times larger” clusters of the future.
Emergence Of A Multi-Model Strategy
The changes are indicative of Microsoft’s changing strategy for AI collaborations. At the town hall, Microsoft CEO Satya Nadella stressed that the company will “certainly support a variety of models” across its products, using GitHub Copilot as an illustration of this multi-model approach.
The auto-selection option provides developers with useful advantages including lower rate limiting and a 10% request discount for paying users. By clicking over chat replies, users can see which model powered each response. The system automatically selects the best model based on performance and capacity.
This change in approach highlights the fierce competition in AI development, where partnership choices are increasingly based on performance metrics rather than conventional vendor ties. Microsoft’s decision to diversify its AI portfolio while keeping its OpenAI investment points to a future in which several AI providers will power various facets of business software applications.