GitHub Copilot drops OpenAI as its default model — Microsoft's own Project Polaris takes over in August
Microsoft is rolling out Project Polaris, its own in-house mixture-of-experts coding model, as GitHub Copilot's default engine through August 2026, replacing GPT-4 Turbo — with a three-month opt-out window before Polaris becomes the only option.
3 August 2026
Microsoft confirmed at Build 2026 that Project Polaris — its own in-house AI coding model, built on a mixture-of-experts architecture and running on Microsoft’s custom Maia accelerators inside Azure — is rolling out through August 2026 as GitHub Copilot’s default engine across every tier: Individual, Business, and Enterprise. It replaces GPT-4 Turbo, which has been Copilot’s default reasoning engine since the product launched. Migration is automatic; teams that want to stay on GPT-4 Turbo get a three-month fallback window before Polaris becomes the only default option.
The technical pitch is Polaris’s per-language tuning — Microsoft says it outperforms GPT-4 Turbo on standard coding benchmarks, with the biggest gains in lower-resource languages like Rust and Haskell, where general-purpose models have historically been weaker. Pro tier subscribers also get multi-file context up to 100,000 lines and autonomous test generation as part of the same rollout.
Why this matters beyond the benchmark numbers
This is Microsoft cutting its highest-volume developer product loose from OpenAI’s models. For most of Copilot’s life, “GitHub Copilot” has effectively meant “GPT running inside VS Code.” That’s no longer true. Microsoft now controls the model, the infrastructure, and the roadmap for the tool a huge share of the industry treats as its coding-assistant default — which changes the calculus for any team whose AI tooling strategy has quietly been “whatever ships as the Copilot default,” rather than a considered choice between Copilot, Claude Code, Cursor, and the rest of the field.
So what
If your team — or your development partner — is running Copilot on autopilot without having actually evaluated it against the alternatives on your specific codebase, this is a natural forcing function to do that comparison now, before the fallback window closes and Polaris becomes the only option. Model choice affects code quality, review overhead, and how much genuine engineering judgment sits behind what gets shipped — it’s not a detail to leave to whichever vendor happens to own the IDE. We evaluate and mix models deliberately as part of how we build; see AI-assisted development for how that fits our process, or get in touch to talk through a project.