Microsoft defaults Copilot to GPT-5.6

Microsoft’s internal Copilot change matters less as a product launch than as a management decision about how engineers are expected to work. CNBC reported on August 5 that Microsoft told developers in its CoreAI group to make OpenAI’s GPT-5.6 Sol the default model in GitHub Copilot for staff coding work, according to an internal memo from Executive Vice President Jay Parikh. (cnbc.com) The instruction does not mean GitHub Copilot changed its public default for all customers. (cnbc.com) GitHub had already added GPT-5.6 Sol, Terra and Luna to Copilot on July 9, and its documentation shows Copilot supports models from OpenAI, Anthropic, Google, xAI, Microsoft and others. (github.blog) What Microsoft did was narrower and more revealing: it picked one model for an internal engineering organization and tied that choice to efficiency, standardization and spending discipline. CNBC said the memo was part of an efficiency push inside CoreAI. ### So what exactly changed inside Microsoft? (cnbc.com) Microsoft’s CoreAI group was told this week to default to GPT-5.6 Sol inside GitHub Copilot when doing staff coding work, CNBC reported, citing Parikh’s memo. The reported rationale was not that other models disappeared, but that employees should center their workflow on OpenAI’s top model. (github.blog) GitHub’s own product materials make clear that Copilot still offers multiple model choices. In GitHub’s July 9 changelog, the company described Sol as the highest-reasoning model in the GPT-5.6 family, Terra as the balanced default for everyday coding, and Luna as the faster, cheaper option. (cnbc.com) ### Why is Sol notable if Terra is described as the balanced default? GitHub described GPT-5.6 Sol as best for “complex reasoning over large codebases” and long-running agentic work. That makes Microsoft’s internal choice notable because it did not simply follow GitHub’s product positioning around Terra as the everyday default. (cnbc.com) CNBC reported that Parikh framed the move around getting more value from Microsoft’s OpenAI access. Gate’s pickup of the CNBC report said Parikh told employees that shifting workloads to OpenAI models would help maximize the company’s token investment value. (github.blog) ### Does this mean Microsoft is shutting out rival models? GitHub’s supported-models page shows no such thing. As of August 6, Copilot lists OpenAI models alongside Anthropic Claude models, Google Gemini models, xAI’s Grok and Microsoft’s own MAI-Code-1-Flash, among others. CNBC contrasted that menu with Anthropic’s Claude Code, which is tied to Anthropic models. (github.blog) Microsoft’s internal instruction therefore sits inside a broader multi-model product, even if one Microsoft team was told to standardize on OpenAI’s flagship option. ### What does this say about Microsoft’s relationship with OpenAI? (cnbc.com) Microsoft 365 Copilot had already moved in the same direction. Microsoft said on July 9 that OpenAI’s GPT-5.6 family became available across Word, Excel, PowerPoint, Chat and Cowork, and called GPT-5.6 the preferred model for Microsoft 365 Copilot. (docs.github.com) OpenAI’s GPT-5.6 family also entered public release on July 9 after a U.S. government review, according to CNBC. That timing meant Microsoft could adopt the same model family across both consumer-facing productivity tools and internal developer workflows within weeks. (cnbc.com) ### Why should developers care about an internal memo? The practical point is that large companies are treating coding models as operating tools, not side experiments. Microsoft’s reported instruction was about the default environment for engineers already building software inside the company, not about a demo or pilot. (techcommunity.microsoft.com) The next public reference points are likely to remain GitHub’s Copilot changelog and supported-models documentation, where model availability and positioning are updated, while Microsoft may disclose more through future product posts or executive comments. (github.blog) (cnbc.com 1) (cnbc.com 2)