Microsoft AI dates the page October 27, 2026. The structured timestamp on the page is 15:10 UTC that day. The model is MAI Image 2.7. The opening calls it the latest image editing model and says it launches in preview. The same paragraph gives two ranks. It says the model launches at No. 3 on Arena Edit, and that this makes Microsoft AI the No. 2 lab. It also says the model is No. 2 on Artificial Analysis's image editing leaderboard. Those ranks sit in one paragraph. They are not labeled as one table. The page title says the model debuts above GPT Image 2.5 Flare on Artificial Analysis and at No. 3 on Arena. The title names Flare. It does not say the model leads every system on the list.
[1]
The product description is separate from the scores. The page says a scene can be restyled or relit, text and symbols can be changed while staying integrated into the image, and existing assets can be restored or enhanced. It adds that some edits require an understanding of how objects, lighting, and environments should work together. A later sentence says image editing comes down to control over details: moving a product into a new setting, shifting the lighting and the mood, and reworking individual objects without losing the larger picture. None of those sentences carries a score. They are the company's description of the edit types.
[1]The comparison with 2.6 is stated as a result from Artificial Analysis. The page says MAI-Image-2.7 improves on 2.6 across all 10 use cases and all 7 editing actions that Artificial Analysis evaluated. It says the model reaches the category frontier in enhancement and restoration, in reasoning-based edits, in scene and style, and in text or symbol editing. The strongest lead it names is scene and style, where the model scores 24 Elo ahead of the next-best model. That sentence does not name the next model. The 24-point figure is attached to one category. It is not a gap printed on an overall bar.
[1]The bar chart is a different list. Its heading says MAI-Image-2.7-Preview ranks 2nd of all public models. The note under the chart says the scores are Elo from blind preference votes in the Image Arena. The source line reads Artificial Analysis, AA-Image-Editing v2.1. The chart's accessible label gives the order: GPT Image 2.5 Sunburst (max) at 1199, MAI-Image-2.7-Preview at 1193, GPT Image 2.5 Flare (max) at 1182, then MAI-Image-2.6, Krea 3 (Ultra), and Nano Banana 2.1 tied at 1151, MAI-Image-2.6-Flash at 1144, and GPT Image 2 (high) at 1136. Sunburst at 1199 sits above the preview at 1193. Flare at 1182 sits below it. The title's claim of sitting above Flare matches that order. A claim of sitting above Sunburst does not. The body's Arena Edit rank of No. 3 and the chart's rank of 2nd are both on the page. The page does not say they count the same votes, and it does not say they use the same cutoff. Merging them into one place on one board adds a step the page does not take.
[1]The 24 Elo lead is also absent from that bar chart. The bars are overall scores. The 24-point sentence sits in the category paragraph. The page does not show an equation that turns 1193, 1182, or 1151 into 24. Reading 24 as the gap between 2.7 and Sunburst, or between 2.7 and Flare, would be a calculation the page does not publish. The category sentence stands on its own: scene and style, 24 Elo ahead of the next-best model, with that model left unnamed.
[1]A second chart covers category ranks rather than overall Elo. Its label says the climbs run from MAI-Image-2.6 to MAI-Image-2.7. Reasoning-based edit moves from rank 7 to rank 1, up 6 places. Scene and style edit moves from rank 7 to rank 1, up 6 places. Text or symbol edits move from rank 6 to rank 1, up 5 places. The subtitle calls these the biggest category rank gains to first place, versus the previous generation. The source line says Artificial Analysis Image Arena, evaluated 22 Oct 2026. The article date is October 27, 2026. The page does not explain the days between the evaluation date and the article date, and it does not say whether the bars were refreshed after October 22. The subtitle's phrase is the previous generation. The body names the comparison model as 2.6. The four frontier categories in the body are enhancement and restoration, reasoning-based edits, scene and style, and text or symbol editing. The climb chart prints three of those moves. It does not print a rank change for enhancement and restoration. The page does not say why that category is missing from the climb chart.
[1]The sentence on where to try the model is short. The page says the editing capabilities can be tried on Arena, on Artificial Analysis, and on MAI Playground, and that the launch starts with editing, with text-to-image generation coming later. It does not give a price, a rate limit, an end date for the preview, or a date for text-to-image. It does not say whether the No. 3 rank on Arena Edit and the 2nd-place Elo chart count the same ballots. The picture with this column is the page header: flowers on a blue ground, with the model name set in type. The line of type on the image is the company's wording. The ranks that name a board are in the body.
[1]