Microsoft pushed out MAI-Image-2.5 on May 26, its newest in-house text-to-image model, announcing it on the company's official blog. The model lands at No. 3 on Arena's text-to-image leaderboard, tied with Google's Nano Banana 2 and behind OpenAI's gpt-image-2, which holds the top spot with a score of 1388.
Here's the catch: No. 3 is the same spot the previous model took back in March. The new version holds the line rather than climbing it. Microsoft's claimed gains sit elsewhere, mostly in text rendering, stylized illustration, and commercial work like product shots and packaging mockups.
"It's our strongest image model yet," per Microsoft AI chief Mustafa Suleyman, who called it a step change in quality. Standard launch language, and the leaderboard parity complicates it. The concrete part: sharper in-image text, steadier layouts, and better handling of lighting, scale, and spatial relationships, all per Microsoft's own description.
It's live on Arena now for anyone to test. Microsoft says it reaches MAI Playground and Foundry within two weeks. The timing isn't random, Build kicks off about a week out.
Bottom Line
MAI-Image-2.5 debuted at No. 3 on Arena, the same rank its predecessor held in March, with OpenAI's gpt-image-2 leading at 1388.
Quick Facts
- Launched May 26, 2026 by Microsoft AI
- Ranked No. 3 on Arena text-to-image leaderboard
- OpenAI's gpt-image-2 leads with Arena score of 1388
- Tied with Google's Nano Banana 2
- Reaches MAI Playground and Foundry within two weeks
