Microsoft tells engineers to default to OpenAI’s GPT-5.6 Sol

Subject: An internal model choice highlights the gap between ecosystem breadth and a practical default.

Preview: Microsoft employees are being directed toward GPT-5.6 Sol in GitHub Copilot, with extra governance for large token-intensive jobs.

A catalogue with thousands of models is useful, but productivity often improves when teams have a well-supported default.

CNBC reports that Jay Parikh told Microsoft engineers to use OpenAI’s GPT-5.6 Sol as the default in GitHub Copilot. Large token-intensive jobs require a conversation with a manager, even as Azure promotes a broad model marketplace.

A default reduces evaluation overhead and concentrates support, prompting guidance and security controls. It also reveals which model Microsoft currently trusts for everyday engineering work.

Default does not mean exclusive or universally best. Different codebases, latency needs and data rules may justify another model, while token controls show that cost still matters.

Set an approved default, document exceptions and route unusually large jobs through lightweight review. Compare productivity, defects and total spend rather than token usage alone.

Model choice at scale is a governance decision: standardise the common path, but keep evidence-based exceptions.