The engine inside the world’s most widely used AI coding assistant has changed, and most developers did not get a memo. Microsoft’s Project Polaris, its first in-house coding model, is replacing GPT-4 Turbo inside GitHub Copilot. For years Copilot ran on OpenAI’s models. Now Microsoft has built its own, and the swap says as much about the OpenAI relationship as it does about the code.
Why Microsoft built its own
Two reasons, and neither is subtle. Cost: running someone else’s frontier model for millions of Copilot users is enormously expensive, and owning the model cuts the bill. Control: depending on OpenAI for the core of a flagship product is a strategic weakness for a company that would rather not have one, especially as its partner grows into a rival. Polaris is Microsoft reducing its reliance on a supplier it also competes with.
What it means for developers
In the short term, probably not much you will feel day to day; Copilot still autocompletes and chats. The thing to watch is quality drift, in either direction. A model tuned in-house for Copilot’s exact workload could get better at the specific job, or a first-generation effort could stumble against a mature GPT model. Keep an eye on whether your suggestions get sharper or vaguer over the coming weeks, and remember you are not locked in.
The take
The interesting story is not the model, it is the divorce it hints at. When your biggest partner starts replacing your technology at the heart of its products, the relationship has entered a new phase. Microsoft building Polaris is a hedge against the day it no longer wants to depend on OpenAI at all, and every developer on Copilot is now part of that hedge, whether they clocked it or not.
Related on Top Tool Stack: Best AI Coding Assistants in 2026 · Cursor Is a SpaceX Company Now
Did you know: GitHub Copilot has been one of OpenAI’s largest showcases for years. Microsoft replacing its engine with a home-grown one is the sort of move partners make right before they stop being partners.