Microsoft used its Build 2026 keynote on June 2 to introduce Project Polaris, an in-house mixture-of-experts coding model that will replace GPT-4 Turbo as the default engine for every GitHub Copilot subscriber starting in August. The switch closes a chapter that began in 2021, when Copilot first shipped on OpenAI Codex, and gives Microsoft end-to-end control of the developer tool with the largest install base.

How to Integrate This

If you pay for GitHub Copilot today, the default will auto-migrate to Polaris in August with no action required. Microsoft is offering a three-month fallback window where teams that need GPT-4 Turbo can pin to it before the legacy model is retired. Build the migration window into your codebase regression tests now: prompt-driven golden traces, suggested-edit acceptance rates per language, and any custom prompt prefixes you ship to your engineers. Polaris uses language-specific tuning, so Python and TypeScript users may see noticeably different defaults than C++ or Rust teams.

Why It Matters

Polaris is the second Microsoft owned-model swap in two months. GitHub already moved to usage-based billing, and pairing that with Microsoft's own coding model removes the OpenAI licensing line item from Copilot's economics. According to Build coverage, Polaris runs on Microsoft's custom Maia AI accelerators in Azure, which should compress inference cost further. The broader signal: Microsoft now intends to compete on coding model quality, not just on developer-tool distribution.

Key Details

  • Architecture: Mixture-of-experts with language-specific routing; built and trained in-house.
  • Default switch: August 2026 for all Copilot subscribers, automatic migration, three-month GPT-4 Turbo fallback option.
  • Infrastructure: Runs on Microsoft's custom Maia 200 AI accelerators in Azure.
  • Multi-agent VS Code mode: Available at Build today, spawns parallel subagents for linting, testing, documentation, and security review via an orchestrator that delegates to specialists.
  • Strategic context: Lands two months after Microsoft and OpenAI ended their seven-year exclusive partnership in April 2026.

What to Do Next

Run Polaris against your real codebase the moment Copilot exposes a model picker. Pin a representative repo to Polaris in one IDE and GPT-4 Turbo in another, then compare acceptance rate, hallucinated import rate, and per-language quality across a week of work. Teams with heavy regulatory or vendor-lock-in concerns should also use the three-month fallback to negotiate Polaris terms before the August default switch is permanent.