← Back to News
ANALYSIS

Microsoft Launches MAI Models at Build 2026, Cutting Copilot's OpenAI Cord

At Build 2026, Microsoft launched its in-house MAI model family — including Project Polaris, a coding model that replaces GPT-4 Turbo as the default for all GitHub Copilot subscribers in August, ending OpenAI dependence.

By Michael Eakins•• min read
MicrosoftMAIGitHub CopilotOpenAIfoundation models

Microsoft opened its Build 2026 developer conference in San Francisco today with the most consequential announcement of its post-OpenAI era: a family of homegrown AI models, collectively branded MAI, that the company will use to run its own products end to end — starting with the one developers touch most, GitHub Copilot.

The headline model is a coding system internally called Project Polaris. According to reporting on internal Microsoft communications, Polaris performs "at or above" Anthropic's Claude 3.7 Sonnet on SWE-bench Verified — the industry standard coding benchmark — while running at significantly lower inference cost. More important than the benchmark is what Microsoft plans to do with it: Project Polaris will replace GPT-4 Turbo as the default engine for all Copilot subscribers starting in August, and it is designed to run natively on Azure without a single OpenAI API call.

What Was Announced

The MAI family launched today spans more than the coding model:

  • Project Polaris — the MAI coding model, benchmarked at or above Claude 3.7 Sonnet on SWE-bench Verified, becoming Copilot's default in August.
  • MAI-Transcribe-1 — a speech-to-text model.
  • MAI-Voice-1 and MAI-Image-2 — voice and image models, made broadly available to commercial developers for the first time.
  • A production-ready Agent Framework for .NET and Python.
  • A rebuilt, multi-model Copilot platform that still offers Anthropic (and other) models as options alongside Microsoft's own.
  • Major Azure AI Foundry updates and Windows-native agent capabilities.

Why It Matters

This is the clearest signal yet that Microsoft intends to compete on model quality, not just distribution. The announcement lands months after Microsoft and OpenAI ended their seven-year exclusive partnership in April 2026, and it gives Microsoft end-to-end ownership of its most widely used developer product.

The motive is as much financial as strategic. Routing every Copilot completion through OpenAI's API made the product's unit economics uncomfortable at enterprise scale — Microsoft was paying a margin to a partner on every keystroke of its own flagship tool. Owning the model and running it on Azure collapses that cost. It is the same vertical-integration logic that Google has already pursued with Gemini across its developer surfaces.

Notably, Microsoft is not ripping third-party models out. The rebuilt Copilot remains multi-model, keeping Anthropic and others available. The change is the default: the model that runs when a developer does nothing special is now Microsoft's, not OpenAI's. Defaults are where the volume — and the cost — live.

What To Watch

The real test arrives in August, when Polaris becomes the default and millions of developers who never change a setting begin running on it. Two questions decide whether this is a milestone or a stumble: does Polaris hold up on the messy, real-world coding tasks that benchmarks don't capture, and does Microsoft keep it the default if quality complaints mount? A quiet revert to an OpenAI model would say more than today's keynote did.

We unpack the strategic stakes — for Microsoft, OpenAI, and every other hyperscaler weighing whether to build its own — in our companion analysis, Microsoft is building its way off OpenAI.

Sources