Microsoft releases MAI-Image-2.5 with better text and visual reasoning

Microsoft Research launched MAI-Image-2.5 on May 26, its strongest image generation model in the MAI-Image series, which now ranks No. 3 on the Arena text-to-image leaderboard. The model is described as 'closer to commercial use,' covering multiple styles and better understanding user instructions. Compared to the previous MAI-Image-2, the new model focuses on improved text rendering, making it suitable for infographics, posters, and packaging. Its visual reasoning is also enhanced, with better object structure, lighting, and spatial relationships. Users can try it on Arena now, and it will be available on MAI Playground and Foundry within two weeks.