Microsoft AI released MAI Image 2.7 on October 27, 2026, and it is an image editing model first. You give it an existing picture and describe the change. Microsoft says it can restyle or relight a whole scene, change text and symbols inside an image, and restore or enhance older assets. Text-to-image generation, the "type a prompt, get a picture" mode, is not part of this release. Microsoft says it is "coming soon."
The launch matters for two reasons. First, the scores Microsoft published put its model within a few points of the top editing model on one public leaderboard. Second, editing is the part of image AI that most everyday tools need: swapping a product's background, fixing a sign, or changing the light in a photo without redrawing everything else.
What MAI Image 2.7 does
Microsoft describes four kinds of edits it is aiming at:
- Scene and style. Restyle or relight a full scene, such as shifting the mood or time of day.
- Text and symbols. Change words or signs in an image while keeping them integrated with the surface they sit on.
- Restoration and enhancement. Clean up or improve existing images.
- Reasoning-based edits. Changes that need the model to understand how objects, lighting and environments work together. Microsoft's example is moving a product into a new setting.
Microsoft also says the model is good at reworking individual objects "without losing the bigger picture." That is the core promise of any editing model: change one thing and leave the rest alone.
How it scored
Microsoft's launch post leans on two outside leaderboards. It says MAI Image 2.7 launches in preview at No. 3 on Arena's image editing board, which makes Microsoft AI the No. 2 lab there, and at No. 2 on Artificial Analysis's image editing leaderboard.
A chart in the post, which cites Artificial Analysis's AA-Image-Editing v2.1, lists these Elo scores:
- GPT Image 2.5 Sunburst (max): 1,199
- MAI-Image-2.7-Preview: 1,193
- GPT Image 2.5 Flare (max): 1,182
- MAI-Image-2.6: 1,151
- Krea 3 (Ultra): 1,151
- Nano Banana 2.1: 1,151
- MAI-Image-2.6-Flash: 1,144
- GPT Image 2 (high): 1,136
So on this board, the new model sits 6 points behind OpenAI's GPT Image 2.5 Sunburst, 11 points ahead of GPT Image 2.5 Flare, and 42 points ahead of Microsoft's own previous model, MAI-Image-2.6. Elo scores come from blind preference votes, where people pick the better of two results without knowing which model made each one. A 6-point gap is small, and leaderboard positions shift as more votes come in.
Microsoft also says MAI-Image-2.7 improves on 2.6 across all 10 use cases and all 7 editing actions that Artificial Analysis evaluates. It claims the model reaches the top of the category in four areas: enhancement and restoration, reasoning-based edits, scene and style, and text or symbol editing. Its largest lead, according to Microsoft, is in scene and style, where it is 24 Elo ahead of the next-best model.
A second chart in the post, citing Artificial Analysis's Image Arena and dated October 22, 2026, shows how far the model climbed in three categories. Reasoning-based edits went from rank 7 under MAI-Image-2.6 to rank 1. Scene and style edits also went from rank 7 to rank 1. Text or symbol edits went from rank 6 to rank 1.
These are Microsoft's own summaries of third-party leaderboards. They are useful for direction, but they are not a substitute for testing your own images.
Where you can try it
Microsoft says you can try MAI Image 2.7's editing today on Arena, Artificial Analysis and MAI Playground. The launch post does not mention Copilot, Bing, PowerPoint or an API.
That last point matters for businesses. Microsoft's developer documentation for MAI image models in Microsoft Foundry, its cloud platform for AI models, lists MAI-Image-2.6, MAI-Image-2.6-Flash and the MAI-Image-2.5 family, all in preview. MAI Image 2.7 is not on that list. OrcaRouter, an API reseller, also reported that the launch came with no Foundry deployment and no published price. If you need an editing model inside an app today, the Foundry list above is what Microsoft offers.
Earlier MAI image models did reach Microsoft's consumer products. When Microsoft announced MAI-Image-2 in Foundry, it said the model was powering image generation in Copilot, with phased rollouts underway in Bing and PowerPoint. Microsoft has not said whether 2.7 will follow the same path.
Why it matters
For creators and marketers, a strong editing model is often more useful than a strong generator. Most real work starts with an image you already have: a product shot, a brand photo, a flyer with a typo. The categories Microsoft highlights, relighting, text changes and restoration, are the edits that older tools tend to get wrong.
For the image AI market, the scores show Microsoft's in-house team now competing near the top of editing leaderboards alongside OpenAI, Google's Nano Banana and Krea. Microsoft builds these models itself, separate from its partnership with OpenAI, and editing is the area where it chose to launch first.
What to watch next: when text-to-image generation arrives, whether MAI Image 2.7 shows up in Foundry with pricing, and whether it reaches Copilot and Microsoft 365 apps. Until then, the free playground is the easiest way to see if the leaderboard gains hold up on your own photos.
For more on the models it is measured against, see our GPT Image 2.5 vs Nano Banana 2.1 comparison and our Krea 3 explainer.
Sources
- Microsoft AI: MAI Image 2.7 debuts above GPT Image 2.5 Flare on Artificial Analysis and #3 on Arena
- Microsoft Learn: Deploy and use MAI image models in Microsoft Foundry
- Microsoft AI: Announcing 3 new world class MAI models, available in Foundry
- OrcaRouter: MAI-Image-2.7-Preview, Six Elo From No. 1, No API Yet
- Artificial Analysis image editing leaderboard
