Microsoft’s MAI-Image-2 enters the top three AI image generators in the world - The Next Web

Microsoft has shipped an image model that sits at number three on the Arena.ai leaderboard — behind only Google Gemini 3.1 Flash and OpenAI GPT-Image 1.5. That result, published this week by Microsoft's AI Superintelligence team led by Mustafa Suleyman, is beginning to arrive across Microsoft's consumer products: MAI-Image-2 is rolling out across Copilot and Bing Image Creator, alongside the company's own MAI Playground. The shift — from depending on OpenAI's models to shipping an in-house alternative — is the point.

MAI-Image-2 is a second-generation model focused on photorealism, readable in-image text, and detailed scene generation. Enterprise API access is available today for select Microsoft customers, with broader developer access through Microsoft Foundry promised soon. The team confirmed its GB200 Blackwell-architecture compute cluster is now operational.

The Arena.ai leaderboard uses crowd-sourced Elo voting and has recorded roughly 4.3 million votes across submitted images. Rankings reflect aggregate human preference rather than fixed-benchmark performance — a distinction that matters when comparing across providers.

The model landed days after Microsoft announced a leadership restructuring that placed Suleyman directly in charge of a new superintelligence team, removing him from day-to-day oversight of Copilot. The reshuffle was publicly announced Tuesday, March 17, 2026, per GeekWire, CNBC, and the Times of India. How immediately the team shipped a visible product is unclear from the announcement itself.