Microsoft released MAI-Image-2.7-Preview on October 27, with AI CEO Suleyman declaring that MAI-Image has risen to second place globally on the Artificial Analysis image editing leaderboard, behind only OpenAI's GPT Image 2.5 Sunburst (max), leaving Google, Meta, xAI and OpenAI's other tier GPT-Image-2.5-Flare far behind.
It is not text-to-image but controllable editing—feed it an original image plus text instructions, and it outputs a revised image. It excels at "harder-to-get-right" tasks: overall style changes, relighting, modifying text and symbols in images, repairing and enhancing material, and complex edits that require understanding the relationships between objects, light and environment.
Four capabilities reach the frontier
Compared with 2.6, 2.7 improves on all 10 use cases and 7 categories of editing actions on Artificial Analysis, with four categories—enhancement and repair, reasoning-based editing, scene style, and text and symbols—reaching the category frontier. Scene and style show the largest lead, 24 Elo above the next-best model; on the AA-Image-Editing v2.1 leaderboard, it ranks second with 1193 Elo.
Users can try it on Arena, Artificial Analysis and MAI Playground; currently only image editing is available, with text-to-image merely stated as "coming soon" with no date given.
As a Preview, no public API, pricing or commercial release date has been announced; the model weights are not open and it cannot be self-hosted, and Playground regional availability may also be limited. To shake OpenAI's top spot, it will still need text-to-image to fill the gap.