The biggest talking point at Microsoft Build on June 2, 2026 was not yet another chat assistant, it was Microsoft formally unveiling its in-house coding model, Project Polaris, and announcing it will replace GPT-4 Turbo as the default model for all GitHub Copilot subscribers starting August 2026. That single step effectively severs Copilot dependence on OpenAI models.
1. Built for Software Development
Polaris is not a general chat model with a new coat of paint. It uses a Mixture-of-Experts architecture with specialized sub-modules for distinct languages, frameworks, and paradigms, and at inference time it adds chain-of-thought and tree-of-thought search to tackle complex multi-file refactors. Microsoft claims it beats GPT-4 Turbo on benchmarks like HumanEval and MBPP, with notable gains in low-resource languages such as Rust and Haskell. It is designed to cover the full development cycle: code generation, multi-file refactoring, test writing, code review, documentation, and dependency analysis.
2. Running on Microsoft Silicon
Polaris runs on Microsoft custom Maia 200 AI accelerators in Azure, which Microsoft says cuts per-inference latency and cost versus routing through OpenAI. Migration is automatic with an optional three-month fallback. The same keynote also brought a multi-agent VS Code and a standalone GitHub Copilot app. Billing changed too: as of June 1, Copilot moved to usage-based GitHub AI Credits, and Copilot code review now consumes Actions minutes.
My Take
The weight here is not "Microsoft built a model", it is "the platform owner is now building its own engine". When the tool developers depend on daily is powered by a model the platform fully controls, bargaining power, cost structure, and data governance all get reshuffled. Two practical lessons for outsourcing teams and enterprise IT: first, never treat any single vendor model as permanent infrastructure, design workflows to be engine-swappable. Second, now that usage-based billing is live, AI cost needs governance like any cloud bill. Uber reportedly burned through its entire annual AI-coding budget in four months, which is less a scare story than a warning. Where AI is used, how much, and who approves it should start entering project cost management, rather than letting engineers spend whenever they feel like it.
Sources
- GitHub Copilot Replaces GPT-4 With Project Polaris — TechTimes
- Microsoft Build 2026 Recap: Project Polaris Cuts the OpenAI Cord — ChatForest
- Microsoft Targets Claude Code with Project Polaris — AI Weekly
- AI News Today — June 2, 2026 — BuildFastWithAI