Felix Laurent
Microsoft AI says MAI Image 2.7 Preview is now #3 on Arena's image editing leaderboard. It also says the model is #2 by lab and ahead of GPT-Image-2. The post puts the gain at 45 points over MAI Image 2.6.
What Arena reported
Arena posted the same result on Oct 27. It says the preview landed at #3 overall in the Image Edit Arena with 1470 points. That makes Microsoft AI #2 among labs, according to Arena.
Arena also says the release is a 45-point improvement over MAI-Image-2.6, which sat at #7. The jump therefore moved the model four places.
Microsoft's own news page repeats the placement. It says the model launches in preview at No. 3 on Arena Edit.
What MAI Image 2.7 can do
MAI Image 2.7 is an image editor, not a full generator yet. Microsoft's launch post says this preview handles edits only, with text-to-image generation "coming soon".
The post lists four kinds of work. The model can restyle or relight a whole scene, change text and symbols inside an image, restore and enhance old assets, and make edits that depend on how objects, light and surroundings fit together. Microsoft's examples include moving a product into a new setting, shifting the mood of a shot and reworking single objects.
It is the fifth model in the line. Microsoft released MAI-Image-1 in October 2025 and MAI-Image-2 in March 2026, followed by versions 2.5 and 2.6 over the summer. All of them are built in-house by Microsoft AI.
How it scores on Artificial Analysis
Arena is not the only board Microsoft cites. The launch post says the model is #2 on the Artificial Analysis image editing leaderboard and debuts above GPT Image 2.5 Flare there.
Microsoft says 2.7 beats 2.6 in all 10 use cases and all 7 editing actions that Artificial Analysis tests. It claims the top spot in four actions: enhancement and restoration, reasoning-based edits, scene and style, and text or symbol editing. Its widest lead is in scene and style, at 24 Elo over the next-best model.
A summary of the board by OrcaRouter gives the overall numbers. It lists 2.7 at 1,193 Elo from 4,891 samples, six points behind OpenAI's GPT Image 2.5 Sunburst (max) at 1,199 and 42 points above MAI-Image-2.6. A six-point gap is small enough to change as more votes come in.
Where to try it
The preview is open in MAI Playground, on Arena and on Artificial Analysis. Microsoft has not announced a public API, a price or a Microsoft Foundry listing for 2.7, and Artificial Analysis marks it as having no API.
The previous version sets a reference point. MAI-Image-2.6 is in Foundry preview at $5 per million text input tokens, $8 per million image input tokens and $38 per million image output tokens, according to OrcaRouter. MAI Playground has historically been limited to US users.
What is not confirmed
The claim about GPT-Image-2 comes from Microsoft's post. Arena's post does not name GPT-Image-2, and we did not open the live leaderboard page.
A Krea benchmark post offers only indirect context. It carries Arena snapshots from Oct 19, but those are for other boards, such as Multi-Image Edit and Single-Image Edit. They are a different snapshot and different boards from the 1470 Arena cites, so they do not confirm the comparison.
Arena's votes also keep arriving. Arena marks early models as preliminary, so a preview rank can shift as the vote count grows.
What it means
The numbers point to Microsoft's in-house image model closing on OpenAI's in editing. It sits #3 on Arena and #2 on Artificial Analysis, six Elo behind GPT Image 2.5 Sunburst on the second board.
That does not make it the best image editor yet. Both ranks rest on preview votes, OpenAI still leads overall, and with no API or price nobody outside the playgrounds can test it on real work. What the release does show is pace: one version moved Microsoft from #7 to #3 on Arena.
