Microsoft’s MAI Image 2.7 Preview Ranks in Top Three for Image Editing

Microsoft AI, the company’s in-house model lab, has released a preview of MAI Image 2.7, an image editing model that has debuted third on Arena’s image editing leaderboard and second on a rival ranking run by Artificial Analysis.

Microsoft describes the model, announced on 27 October, as “our latest image editing model”, with “even stronger control for edits that are hardest to get right”. For now it only edits existing pictures. In its announcement the lab says that “text-to-image generation” is “coming soon”. The model is available to try on Arena, Artificial Analysis and Microsoft’s own MAI Playground.

What the model does

Microsoft says MAI Image 2.7 can “restyle or relight an entire scene, change text and symbols while keeping them integrated into the image, or restore and enhance existing assets”. It also pitches the model at edits that require understanding how objects, lighting and surroundings fit together, such as moving a product into a new setting.

The lab says the model improves on MAI-Image-2.6 in every category Artificial Analysis measures, and that “its strongest lead is in scene and style, where it scores 24 Elo ahead of the next-best model”. Elo is a rating system, borrowed from chess, that turns head-to-head wins and losses into a single score.

Where it ranks, according to the leaderboards

Both leaderboards rank models using blind preference votes: people compare edited pictures without knowing which model made them. The rankings below come from the two companies that run them.

Arena, which raised $200 million this month to rank AI models, said in a post on X that MAI-Image-2.7-Preview “just landed at #3 overall in the Image Edit Arena with 1470 pts! That puts Microsoft AI at #2 among labs.” It called the result “a +45 pt improvement” on MAI-Image-2.6, which sat seventh. Arena said the new model ranks third for portraits, for product and branding design and for photorealistic images, and fourth for text rendering.

Artificial Analysis said in its own thread that the model “takes #2 on AA-Image-Editing v2.1, close behind OpenAI’s GPT Image 2.5 Sunburst (max)”. A chart in Microsoft’s post, citing that leaderboard, puts the gap at 1,193 to 1,199, with OpenAI’s GPT Image 2.5 Flare third at 1,182. Artificial Analysis said MAI Image 2.7 ranks first on four of the seven editing actions it measures, but that the OpenAI model still scores higher on composition, editing individual objects and keeping a person’s identity intact.

How Microsoft is selling it

Mustafa Suleyman, who runs Microsoft AI, wrote on X: “MAI-Image is now the #2 image edit model in the world!”, adding that it was “once again beating Google, Meta, xAI, and GPT-Image-2.5-Flare”. That claim matches the Artificial Analysis ranking. On Arena’s overall table the model is third, with Microsoft second among labs.

The Microsoft AI account highlighted the Arena result, saying the model is “#2 by lab and ahead of GPT-Image-2”.

Why leaderboard ranks have become the launch pitch

Microsoft has built this launch around other people’s leaderboards rather than its own benchmarks, and that is a sensible choice. Blind voting by the public is harder to game than a test a lab designs for itself, and two separate leaderboards agreeing that the model sits near the top is stronger evidence than either alone. It is a contrast with the lab’s MAI-Base-2 announcement earlier this month, which rested on Microsoft’s own private tests.

The rankings still have limits. Preference votes show which picture people like more, which may not capture whether an edit followed a detailed brief exactly, and ratings can shift as more votes arrive. The model is also a preview, without pricing or a stated route into Microsoft products such as Copilot.

The bigger signal is competitive. Image editing has become a crowded field in recent weeks, with Krea 3, Bria’s open-weight Fibo 2.0 and Canva’s Craft all arriving this month. Microsoft is now placing its own in-house model just behind OpenAI’s best on one public chart, and ahead of Google, Meta and xAI on that ranking, according to Suleyman.

Sources

1. Microsoft AI, “MAI Image 2.7 debuts above GPT Image 2.5 Flare on Artificial Analysis and #3 on Arena”, 27 October 2026. https://microsoft.ai/news/mai-image-2-7-debut/
2. Microsoft AI (@MicrosoftAI) on X, 27 October 2026. https://x.com/MicrosoftAI/status/2115102961957499069
3. Mustafa Suleyman (@mustafasuleyman) on X, 27 October 2026. https://x.com/mustafasuleyman/status/2115099284224008200
4. Arena (@arena) on X, 27 October 2026. https://x.com/arena/status/2115101113666134276
5. Artificial Analysis (@ArtificialAnlys) on X, 27 October 2026. https://x.com/ArtificialAnlys/status/2115097540320506069

Related stories

• Microsoft AI Finishes Pre-Training MAI-Base-2, a 2.8-Trillion-Parameter Model, and Starts Preparing MAI-Base-3
• Krea Launches Krea 3 Image Editing Model for Sub-8-Second 2K Edits
• Bria’s Fibo 2.0 Leads Open-Weights Image Editing, Under a Non-Commercial Licence
• Canva Releases Craft, Its First In-House Image Generation Model
• Arena Raises $200M at $3.1bn Valuation to Rank AI Models