Image: Microsoft AI
Microsoft AI has released MAI-Image-2.7-Preview, a new image editing model, and it has entered Arena's Image Edit leaderboard in third place. Microsoft announced the model on October 27, calling it its latest image editing model with stronger control over the edits that are hardest to get right.
MAI-Image keeps climbing!
— Microsoft AI (@MicrosoftAI) October 27, 2026
MAI Image 2.7 Preview is now #3 on @arena's image editing leaderboard. #2 by lab and ahead of GPT-Image-2. That's a +45 point jump from 2.6.
The model is built for changing pictures that already exist rather than making new ones from a text prompt. According to Microsoft, it can restyle or relight a whole scene, change text and signs inside an image while keeping them natural, and restore or enhance older assets. Microsoft says text to image generation is coming soon.
Arena, the public benchmarking site, put the preview at #3 overall with a score of 1,470. Only two OpenAI models sit above it, GPT Image 2.5 Sunburst at 1,515 and GPT Image 2.5 Flare at 1,483. That makes Microsoft AI the second best lab on the board, ahead of GPT Image 2 Medium at 1,460, Grok Imagine Image 2.0 at 1,439 and Google's Gemini Nano Banana 2.1 at 1,427.
Exciting news: MAI-Image-2.7-Preview from @MicrosoftAI just landed at #3 overall in the Image Edit Arena with 1470 pts! That puts Microsoft AI at #2 among labs.
— Arena.ai (@arena) October 27, 2026
This release is a +45 pt improvement from its previous variant, MAI-Image-2.6 at #7.
The jump over the previous version is the bigger story. MAI-Image-2.6 sits at #7 with 1,425, so the new model gained 45 points and climbed four places. Arena says it improved in every category it tracks. It now ranks third in product and branding, portraits and photorealistic images, fourth in text rendering and in cartoon and anime, and fifth in 3D. Text rendering saw the largest move, up from 11th.
Arena scores come from people, not automated tests. Visitors submit an editing request, see results from two anonymous models side by side and vote for the better one. Those votes are turned into a rating, so a higher score means a model wins more of these blind matchups. Scores can shift as more votes come in, and the preview's margin of error on the chart is wider than that of older entries.
A second benchmark points the same way. Microsoft says the model ranks #2 on Artificial Analysis's image editing leaderboard and improved on 2.6 in all ten use cases and all seven editing actions there. The company claims a 24 point lead over the next best model in scene and style edits.
For regular users, the practical upshot is better results when asking for things like swapping a product's background, changing the lighting of a photo or fixing text on a sign. Microsoft says people can try the editing feature today on Arena, Artificial Analysis and its MAI Playground. API reseller OrcaRouter reports that there is no API, Microsoft Foundry listing or published price for 2.7 at the time of this writing.
Since it is still a preview, its ranking and features may change before a full release.
Comments