Microsoft Unveils Project Polaris — Its Own AI Model to Replace GPT-4 in GitHub Copilot

Microsoft Build 2026 opened June 2 at Fort Mason Center in San Francisco with what may be the most consequential developer tooling announcement of the year. Satya Nadella unveiled Project Polaris — Microsoft's in-house AI coding model — which will replace GPT-4 Turbo as the default engine powering GitHub Copilot for all Individual, Business, and Enterprise subscribers starting August 2026.

Polaris is a mixture-of-experts architecture with specialized sub-modules tuned for different programming languages and frameworks. Microsoft claims it outperforms GPT-4 Turbo on HumanEval and MBPP benchmarks, with particular gains in low-resource languages like Rust, Haskell, OCaml, Elixir, and Zig. Pro-tier subscribers will gain access to multi-file context up to 100,000 lines and autonomous test generation capabilities.

The model runs on Microsoft's custom Maia AI accelerators inside Azure, reducing per-inference latency and lowering operational cost. Enterprise customers get an optional three-month fallback period to stay on GPT-4 Turbo during the transition.

The significance extends beyond benchmarks. With nearly 140,000 organizations and 4.7 million paid subscribers on Copilot, this is the highest-volume AI coding deployment in the world transitioning away from OpenAI infrastructure. Microsoft now controls the model, the inference infrastructure, and the developer experience end-to-end. The Copilot SDK — in public preview since April 2026 — ships a reasoning layer built entirely within Microsoft's stack.

Build 2026 also delivered the Windows Agent Framework as open-source under MIT license, Azure Agent Mesh for federated agent execution, and the Windows Agent Store with an 85% revenue share for developers selling agent manifests.