Microsoft is adding an automatic AI model selector to Visual Studio Code that will choose the best-performing model for its GitHub Copilot service.
The new feature will prioritize Anthropic’s Claude 4 Sonnet over OpenAI’s GPT-5 for most users, according to company announcements and internal directives.
How the AI model selector works
The selector will automatically evaluate and choose from a range of models, including Claude 4 Sonnet, GPT-5, and GPT-5 mini. Users of the free version of GitHub Copilot will be able to manually select from the available models. However, for paid subscribers, the system will default to Claude 4 Sonnet, indicating Microsoft’s preference for the model in its developer tools.
This move is consistent with Microsoft’s internal strategy. Sources familiar with the company’s plans said Microsoft has been directing its own developers to use Claude 4 Sonnet for several months.
The guidance originated from a June email sent by Julia Liuson, head of Microsoft’s developer division.
“Based on internal benchmarks, Claude 4 Sonnet is our recommended model for GitHub Copilot.”
This recommendation has remained in place even after the release of OpenAI’s more advanced GPT-5 series, as subsequent testing confirmed Claude 4 Sonnet’s superior performance on benchmarks specific to GitHub Copilot tasks.
Microsoft’s broader AI strategy and in-house development
Microsoft is also integrating Anthropic’s models into its Microsoft 365 Copilot for certain tasks in Excel and PowerPoint, where they reportedly outperformed OpenAI’s models during testing.
Alongside these partnerships, Microsoft continues to build its own AI capabilities. At a recent employee town hall, Microsoft AI chief Mustafa Suleyman discussed the company’s investment in its own infrastructure. He mentioned that a preview of its MAI-1 model was trained on a relatively small cluster of 15,000 H100 GPUs, signaling plans for much larger clusters in the future.
“We’re also going to be making significant investments in our own cluster. So today, MAI-1-preview was only trained on 15,000 H100s, a tiny cluster in the grand scheme of things.”
The news comes as OpenAI and Microsoft announced a new agreement last week, a deal that could help resolve operational details and facilitate OpenAI’s potential move toward an initial public offering.
