Microsoft releases the MAI-Image-2.7-Preview image editing model, claiming it ranks second worldwide
Microsoft has released the MAI-Image-2.7-Preview image editing model, which focuses on taking an original image plus a text instruction as input and outputting an edited image; text-to-image generation is not yet available. On Artificial Analysis's AA-Image-Editing v2.1 leaderboard, the model ranks 2nd with 1,193 Elo, second only to OpenAI's GPT Image 2.5 Sunburst (max), and leads the next-best model by 24 Elo in scene and style editing. Users can try it on Arena, Artificial Analysis, and the MAI Playground; Microsoft has not yet announced a public API, pricing, or an official commercial launch date, and the model weights have not been released.
Why read this
AI score 64
Microsoft discloses the ranking and lead margin of its new image editing model on the Artificial Analysis leaderboard, allowing comparison of its gap with OpenAI.
Topics
Body · AI translation
"World's No. 2 Image Editing AI Model": Microsoft MAI-Image-2.7-Preview Debuts, Outperforming Google, Meta, and xAI
2026-10-28 10:37 IT Home - Guyuan
IT Home, October 28 news: Microsoft yesterday (October 27) issued an announcement announcing the launch of the MAI-Image-2.7-Preview model. Microsoft AI CEO Mustafa Suleyman proudly announced: "MAI-Image is now the world's No. 2 image editing model!"
In his tweet, Suleyman wrote that, based on the Artificial Analysis image editing leaderboard, Microsoft's MAI-Image series models have once again beaten Google, Meta, xAI, and GPT-Image-2.5-Flare to become the world's No. 2 image editing model, second only to OpenAI's GPT Image 2.5 Sunburst (max). IT Home attaches the relevant screenshot below:
It should be noted that at this stage, the MAI-Image-2.7-Preview model focuses on "input original image + text instruction → output edited image" and has not yet opened up text-to-image functionality; it is a controllable editing model for existing photos, design drafts, screenshots, or film and television materials.
Microsoft emphasizes that it can handle tasks that are "hard to get right": overall style changes, relighting, modifying text/symbols in the image, restoring and enhancing materials, and complex edits that require understanding the relationships among objects, light, and the environment.
Microsoft says that compared with the previous generation MAI-Image-2.6, MAI-Image-2.7 has improved on all 10 test cases and 7 types of editing actions in the Artificial Analysis evaluation, and has reached the category frontier in four categories: enhancement and restoration, reasoning editing, scene style, and text and symbols.
Among them, scene and style editing leads by the largest margin, 24 Elo higher than the next-best model; on Artificial Analysis's AA-Image-Editing v2.1 leaderboard, it ranks No. 2 with 1,193 Elo, second only to OpenAI's GPT Image 2.5 Sunburst (max).
Users can currently try it on Arena, Artificial Analysis, and MAI Playground, and currently only image editing is available. Microsoft says text-to-image is "coming soon" but has not announced a date.
The model is still a Preview version at this stage, and Microsoft has not yet announced a public API, pricing, or an official commercial launch date; the model weights are also not open, so it cannot be self-hosted. Playground regional availability may also be limited.