MAI Image 2.7 is Microsoft AI's image model, released in preview on October 27, 2026. It edits existing images and does not yet generate from text. It ranks #2 on Artificial Analysis's image editing leaderboard and #3 on Arena Edit, behind two OpenAI GPT Image 2.5 variants. You can try it in MAI Playground, Arena and Artificial Analysis, but there is no API or price yet.
MAI Image 2.7 in one paragraph
MAI Image 2.7 is a preview image editing model from Microsoft AI, released on October 27, 2026, that changes images you give it and does not yet generate them from scratch. It debuted at No. 3 on Arena Edit and No. 2 on Artificial Analysis’s image editing leaderboard.
If you saw “MAI Image 2.7” described as a new text-to-image model, that is wrong. Microsoft’s own announcement calls it an editing model and says text-to-image is “coming soon”.
| Fact | Detail |
|---|---|
| Maker | Microsoft AI |
| Released | October 27, 2026, in preview |
| Does | Edits supplied images (restyle, relight, text changes, restoration) |
| Does not do (yet) | Text-to-image generation |
| Artificial Analysis editing | #2, 1195 Elo |
| Arena Edit | #3, about 1,470 (preliminary) |
| Try it | MAI Playground, Arena, Artificial Analysis |
| API and price | Not announced |
Prices and rankings here were checked in November 2026. Both leaderboards move as votes come in.
What MAI Image 2.7 does well
MAI Image 2.7 is built for edits that models usually get wrong. Microsoft names four strengths, and Artificial Analysis’s per-task data backs them up.
- Scene and style. Restyle or relight a whole scene while keeping the subject. Microsoft says its scene-and-style score is 24 Elo ahead of the next-best model
- Text and symbols. Change words or marks inside an image so they stay integrated with the lighting and surface
- Enhancement and restoration. Clean up or improve existing assets
- Reasoning-based edits. Edits that depend on how objects relate to each other in the picture
Microsoft says 2.7 improves on MAI-Image-2.6 across all 10 use cases and all 7 editing actions Artificial Analysis evaluates. Per coverage of the leaderboard from WinBuzzer, the biggest jumps over 2.6 were composition and framing (+57 Elo), text or symbol editing (+49) and scene and style (+48).
The same coverage says 2.7 ranks first in four actions: scene and style, text or symbols, reasoning-based editing, and enhancement and restoration. Treat these as strong signals, not guarantees: the scores come from human preference votes, not from your images.
Where MAI Image 2.7 ranks against GPT Image 2.5 and Google
MAI Image 2.7 ranks second behind OpenAI on Artificial Analysis and third on Arena, and it beats Google’s Nano Banana 2.1 on both. The gap to the top is inside the margin of error on Artificial Analysis.
The AA-Image-Editing v2.1 leaderboard on Artificial Analysis, captured November 6, 2026
Here are the top of the Artificial Analysis image editing leaderboard as of our check:
| Rank | Model | Elo | Released | Listed price per 1,000 images |
|---|---|---|---|---|
| 1 | GPT Image 2.5 Sunburst (max) | 1199 | Sep 2026 | $210.72 |
| 2 | MAI-Image-2.7-Preview | 1195 | Oct 2026 | No API available |
| 3 | GPT Image 2.5 Flare (max) | 1182 | Sep 2026 | $210.72 |
| 4 | Krea 3 (Ultra) | 1152 | Oct 2026 | $100.00 |
| 5 | Nano Banana 2.1 (Google) | 1152 | Oct 2026 | $33.60 |
| 6 | MAI-Image-2.6 | 1149 | Aug 2026 | $38.91 |
Two things stand out. MAI Image 2.7 is about 4 points behind the leader with overlapping error bars, and it is 43 points ahead of Google’s best entry.
On Arena’s Image Edit board, Microsoft says the preview launched at No. 3, which makes Microsoft AI the No. 2 lab. Reported scores put it near 1,470, behind both GPT Image 2.5 variants. Arena notes the scores are preliminary, and 2.7 had far fewer votes than its rivals at launch, so expect it to move.
For a closer look at the models it is chasing, see our posts on GPT Image 2.5 and GPT Image 2.5 vs Nano Banana 2.
What MAI Image 2.7 does not do yet
MAI Image 2.7 has three limits you should know before you plan around it.
- No text-to-image. Microsoft gives no date. MAI-Image-2.6 stays in MAI Playground’s generation selector for now
- No API. Artificial Analysis lists no developer API, and Microsoft has not announced pricing or a Foundry listing
- Preview status. Microsoft calls it a preview, and its Arena vote count is small, so rankings can shift
So you can use it today by hand, but you can’t build a product on it. If you need an editing model through an API now, GPT Image 2.5 and Nano Banana 2.1 both have listed prices on Artificial Analysis.
How to try MAI Image 2.7
You can try MAI Image 2.7 in three places, per Microsoft’s announcement. Pick the one that fits what you want to learn.
- MAI Playground. Microsoft’s web app at playground.microsoft.ai. Upload an image, write the edit as plain text and run it. Coverage reports you need a Microsoft account, and Microsoft has not published usage limits for 2.7
- Arena. Open the Image Edit battle mode on arena.ai. You upload an image and an instruction, get two anonymous results, and vote. 2.7 appears as one of the possible models
- Artificial Analysis. Its image arena runs the same blind-vote format and feeds the leaderboard above
Prompts that suit it
Because Microsoft’s pitch centers on hard edits, test the cases where other tools slip:
- Relight a scene. “Change this daytime street to night, with wet pavement and neon reflections”
- Swap text. Replace the words on a sign or label and check that the font, perspective and lighting still match
- Restore a photo. Upload a scratched or low-quality image and ask for a clean, sharp version
- Edit by reasoning. Ask for a change that depends on context, such as “remove the second person and fill in what would be behind them”
Run the same image through two models in Arena and compare. That tells you more than any leaderboard.
MAI Image 2.7 vs earlier MAI models
MAI Image 2.7 replaces MAI-Image-2.6 for editing, but not for generation. On Artificial Analysis, 2.6 scores 1149 in editing against 1195 for 2.7, a 46-point gap.
Microsoft has been shipping fast: 2.6 was listed in August 2026 and a cheaper 2.6-Flash variant in September, per Artificial Analysis. If your work is generation, stay on 2.6 until the 2.7 text-to-image release lands.
Who should care about MAI Image 2.7
MAI Image 2.7 matters most if you edit existing images for a living or are choosing which editing model to bet on.
- Designers and marketers who retouch product shots, restyle campaign images or fix text in assets
- Teams comparing models who want a third serious option beside OpenAI and Google
- Anyone who wants to wait for the API and price before committing, since neither exists yet
If you want a stylized generator instead, read our Midjourney V8 review. And once an edited still looks right, you can animate it: our guides to the best text-to-video AI, Google Veo alternatives and best open-source video models cover image-to-video options.
Verdict
MAI Image 2.7 is a credible top-three image editor that you can test now but not yet build on. It trails OpenAI by a hair, beats Google’s Nano Banana 2.1 on both leaderboards, and its text-to-image sibling is still unreleased.
Test it on your hardest edit this week. Revisit when Microsoft publishes an API, a price and the generation model.