Microsoft Breaks Free from OpenAI with MAI Models and Autonomous Superintelligence Ambitions
> cd .. / HUB_EDITORIALE
News

Microsoft Breaks Free from OpenAI with MAI Models and Autonomous Superintelligence Ambitions

[2026-06-06] Author: Ing. Calogero Bono
> share
Zenithby Meteora Web The operating system for your business. Social, clients, bookings and invoices in one platform. Gyms, barbers, professionals. Discover Zenith Free demo · no card

Microsoft has officially severed its exclusive reliance on OpenAI. At Build 2026, Microsoft AI CEO Mustafa Suleyman revealed that a contractual renegotiation six months ago freed the company to pursue superintelligence independently. The revised deal removed restrictions that prevented Microsoft from training its own frontier models. The result is the MAI model family, seven in-house AI models designed for reasoning, code, image generation, transcription, and voice synthesis.

MAI Models and the Path to Self-Sufficiency

Flagship MAI-Thinking-1 is a 35-billion-parameter reasoning model trained from scratch on clean, licensed data. Unlike many competitors, Microsoft does not distill from other labs. The suite includes MAI-Code-1-Flash for GitHub Copilot, MAI-Image-2.5 for text-to-image and editing, MAI-Transcribe-1.5 covering 43 languages, and MAI-Voice-2 for multilingual speech. All models ship via Microsoft Foundry and can be fine-tuned by enterprises using Frontier Tuning, a reinforcement learning environment that adapts AI to proprietary workflows.

Sponsored Protocol

From Chatbots to Autonomous Agents

Suleyman introduced the Actions Quotient (AQ) as the next evolution beyond IQ and EQ. Microsoft Scout, the first Autopilot agent, runs as an always-on assistant with governed identity in Microsoft Entra. Foundry now supports agents with sub-100-millisecond cold starts and one-click deployment to Teams and Copilot. Frontier Tuning has already delivered impressive results: a model tuned for Excel matches GPT 5.4 at one-tenth the cost.

Sponsored Protocol

Why Custom Silicon Matters

Microsoft is the world's largest GPU buyer but is building its own chips. The Maia 200 accelerator is 30% more cost-efficient than Nvidia GB200, and when running optimized MAI models, performance per watt improves by another 1.4x. This vertical integration positions Microsoft uniquely in the AI arms race. For more on AI security risks and cyber threats, check our coverage on AI Hacking and Airborne Speaker Hack and IBM Whistleblower.

External source: WIRED - Has Microsoft Lost Its Mojo (Again)?

> share
Ing. Calogero Bono

> AUTHOR_EXTRACTED

Ing. Calogero Bono

Ingegnere informatico, fondatore di Meteora Web e Zenith OS. System administrator e progettista di piattaforme, app e CMS proprietari, con esperienza in sviluppo full-stack, marketing digitale ed ecosistema Google.
[ Read Full Dossier ]

> METEORA_WEB // DIGITAL AGENCY

We build the digital presence your business deserves.

Websites, social media, online advertising, e-commerce and high-performance hosting, engineered with method by computer engineers in Sciacca, for all of Italy.

> MW_JOURNAL

> READ_ALL()