The Pulse
Microsoft Puts MAI Image 2.7 Editing Model Into Preview
Microsoft opened a preview of MAI Image 2.7, an image-editing model for changing scenes, lighting, text and symbols in existing images. The company says it ranks third on Arena Edit and second on Artificial Analysis, with text-to-image gene
AI.info Team · Written with AI · fact-checked ·
Two leaderboards, two places near the top
Microsoft’s MAI Image 2.7 entered preview on October 27, with a third-place debut on Arena Edit and second place on Artificial Analysis’s image-editing leaderboard. The different positions reflect separate evaluations, not a direct disagreement: Microsoft cites both rankings as evidence for the launch, while each board measures preferences among its own comparisons. Artificial Analysis currently lists MAI Image 2.7 Preview at 1,193 Elo, with a rank range of first to third across 4,891 samples.
That score puts it just below OpenAI’s GPT Image 2.5 Sunburst (max), listed at 1,199 Elo. Their 95% confidence intervals overlap, so the narrow point-score gap does not establish a clear lead in editing quality. Artificial Analysis says its rankings draw on blind user preference votes, making them a useful comparative signal rather than a verdict on every creative brief. Artificial Analysis’s leaderboard gives the model no available API listing.
Editing comes before image generation
Microsoft presents 2.7 as a model for changing existing images: restyling or relighting a scene, revising text and symbols so they fit the image, and restoring or enhancing existing assets. Its examples include moving a product into a new setting and altering individual objects while preserving the wider scene. Those tasks require the model to change requested details without losing the relationships among objects, lighting and surroundings.
“We’re starting with editing, with text-to-image generation coming soon,” Microsoft says in its launch announcement. The preview therefore starts with image-to-image work; the company does not give a date for the text-to-image feature.
Microsoft reports gains across its test categories
Microsoft says MAI-Image-2.7 improves on version 2.6 across all 10 use cases and all seven editing actions evaluated by Artificial Analysis. The company identifies enhancement and restoration, reasoning-based edits, scene and style, and text or symbol editing as categories where the model reaches the frontier. Its largest stated advantage is in scene and style, where Microsoft says it scores 24 Elo ahead of the next-best model.
The release does not publish the underlying scores for each use case or editing action, and it does not describe a Microsoft-run comparison outside those evaluations. The category results and improvement claims are Microsoft’s account of the benchmark, not a guarantee that every edit will preserve the details a user wants to keep.
Preview access is public; production terms are not
Microsoft directs users to try MAI Image 2.7 on Arena, Artificial Analysis and MAI Playground. Its announcement frames the release as a preview and focuses on editing rather than developer deployment. It gives no price, API availability or timetable for bringing version 2.7 to Microsoft Foundry.
For now, the concrete offer is an image-editing preview with a strong early placement on two preference boards. Users can test scene changes, object edits, restoration and embedded text changes; text-to-image generation remains a future addition, without a release date.