Skip to content
News Product launch

Microsoft unveils Project Polaris, its own AI model to replace GPT-4 Turbo in GitHub Copilot from August 2026

· by Pondero Newsdesk

The short version

Satya Nadella opened Microsoft Build 2026 on June 2 by announcing Project Polaris, a Microsoft-built mixture-of-experts model that will replace GPT-4 Turbo as the default reasoning engine in GitHub Copilot starting August 2026.

Microsoft unveils Project Polaris, its own AI model to replace GPT-4 Turbo in GitHub Copilot from August 2026

Satya Nadella opened Microsoft Build 2026 in San Francisco on June 2 by announcing Project Polaris, a Microsoft-built mixture-of-experts model set to replace GPT-4 Turbo as the default reasoning engine in GitHub Copilot starting August 2026.

What

Project Polaris is Microsoft's in-house AI coding model, designed specifically for software development tasks. Per the Build 2026 keynote, Polaris uses a mixture-of-experts architecture with sub-modules tuned for different programming languages and frameworks. Microsoft said it outperforms GPT-4 Turbo on HumanEval and MBPP benchmarks, with particular gains in Rust and Haskell (per ChatForest's Build 2026 recap).

The model runs on Microsoft's custom Maia AI accelerators inside Azure. Per Microsoft, this reduces per-inference latency and lowers cost compared to routing Copilot inference through OpenAI's API. The August 2026 rollout includes automatic migration for all Copilot subscribers, with an optional three-month GPT-4 Turbo fallback period for enterprise teams that want to stay on the current model before committing to the switch.

For Pro-tier subscribers, Polaris ships with two new capabilities: multi-file context windows up to 100,000 lines of code, and autonomous test generation. The Copilot SDK (in public preview since April 2026) also switches to Polaris as its default reasoning layer at the same time.

Build 2026 also saw Copilot Workspace graduate from beta to general availability. The GA release adds Copilot Extensions for Jira, Datadog, and ServiceNow, plus Fleet mode for autonomous operation on narrowly scoped tasks and Autopilot mode for scheduled background work on issues without a developer present (per Build 2026 coverage from ChatForest).

Why it matters

The OpenAI-GitHub Copilot arrangement carried structural tension since the two companies compete for developer mindshare. With Polaris, Microsoft now controls the model, the inference infrastructure, and the developer experience. The commercial loop that routed all Copilot inference revenue through OpenAI's API no longer applies to the flagship coding product.

Whether Polaris actually performs better on real codebases than GPT-4 Turbo did is a separate question. HumanEval and MBPP are standard benchmarks, but they measure general coding ability on public-domain problems. Gains on Rust and Haskell are notable because those languages have historically been underrepresented in training data. Real-world gains on TypeScript, Python, and Java, which account for the bulk of Copilot usage, have not been disclosed.

The three-month fallback option signals that Microsoft expects some friction. Enterprise teams using Copilot SDK integrations may see different behavior under Polaris than under GPT-4 Turbo, and the fallback gives those teams a window to test and adjust before the switch is permanent.

GitHub Copilot previously was essentially a productized GPT-4 Turbo. That framing shaped how developers compared it against Cursor and Claude Code, which compete on the quality of purpose-built models. Polaris is Microsoft's claim that Copilot's model can now be evaluated on its own merits rather than as a packaged version of someone else's general-purpose frontier model.

Context and reactions

Build 2026 ran June 2 to 3 at Fort Mason Center in San Francisco with roughly 2,500 developers on-site, per conference coverage. The Polaris announcement came as the keynote's lead item, which signals how Microsoft is prioritizing the message: developer independence from OpenAI is the headline, not a side note.

Alongside Polaris, Microsoft announced the MAI model suite v2, covering image generation (MAI-Image-2.5), voice (MAI-Voice-2), and transcription (MAI-Transcribe-1.5). V2 pricing had not been disclosed as of June 2. The April 2026 v1 pricing was $0.36 per hour for transcription, $22 per million characters for voice, and $5/$33 per million tokens (text output/image output) for image generation (per the ChatForest Build 2026 recap). The coordinated push across three modalities alongside Polaris points to a broader strategy to reduce Microsoft's AI pipeline dependence on OpenAI across all product surfaces, not just Copilot.

OpenAI had not issued a public statement in response to the Polaris announcement as of June 2.

What to watch next

Microsoft has not disclosed whether Polaris will be available outside of Copilot (via Azure AI Foundry or the API) or whether it remains a Copilot-exclusive model at launch. GitHub blog posts with pricing and availability specifics were expected same-day at github.blog. The three-month fallback window means the effective forced-migration deadline for enterprise teams is November 2026.

Sources