Project Polaris: Microsoft Drops GPT-4 From GitHub Copilot and Goes In-House

Microsoft just made its boldest move yet to take back the AI coding race. At Build 2026 in San Francisco today, the company unveiled Project Polaris, an in-house coding model that will replace GPT-4 Turbo as the default engine for all GitHub Copilot subscribers starting August.

Project Polaris: Microsoft's in-house answer to Claude Code

Polaris is built on a mixture-of-experts architecture with specialized sub-modules tuned per programming language. Microsoft claims it outperforms GPT-4 Turbo on HumanEval and MBPP benchmarks, with notable gains in low-resource languages like Rust and Haskell. Those numbers haven't been independently verified yet.

Every Copilot subscriber will be auto-migrated, with an optional three-month fallback to GPT-4 for teams that need breathing room. The model runs on Microsoft's custom Maia AI accelerators inside Azure.

Why Microsoft is cutting the OpenAI cord on Copilot

This announcement comes barely five weeks after Microsoft and OpenAI ended their exclusive partnership on April 27. The license still runs through 2032, but exclusivity is gone. Redmond clearly wants end-to-end ownership of its biggest developer product.

The competitive pressure is real. Copilot's market share among professional developers dropped from 67% to 51% in a year, while Anthropic's Claude Code surged to 41% among agentic coding users. Microsoft even named Claude Code as the specific competitor Polaris is built to displace. When was the last time a Big Tech incumbent publicly admitted who was eating its lunch?

Multi-agent mode, Workspace GA, and the MAI suite

Polaris arrived alongside several other major announcements. Multi-agent mode in VS Code now lets Copilot spawn parallel sub-agents handling linting, testing, docs, and security review simultaneously. Copilot Workspace reached general availability, giving agents the ability to reason across full repositories.

Microsoft also rolled out the MAI v2 model suite covering image generation, voice synthesis, and transcription. This is the first coordinated push to replace OpenAI models across multiple product lines at once.

Can Polaris win back developers from Claude Code and Cursor?

That's the trillion-dollar question. Cursor already crossed $2 billion ARR, and Claude Code hit $2.5 billion in under 10 months. Microsoft still has 26 million users and 90% Fortune 100 penetration, but momentum has shifted.

Polaris gives Microsoft full control of the engine. Whether that engine can compete with Opus 4.8 and its million-token context window is the question August will answer. Independent benchmarks will tell the real story.