MAI-Image-2.6-Preview: #1 in Image Editing, #2 in Text-to-Image

Microsoft's MAI-Image-2.6-Preview is no longer just an Arena leaderboard entry you can't touch — it is now an independent leaderboard-topper. On August 25, Artificial Analysis, the independent benchmarking outfit, ranked the same preview build No. 1 on its Image Editing Leaderboard and No. 2 in Text to Image. The No. 1 editing spot puts it ahead of Microsoft's own MAI-Image-2.5-Pro (the previous leader), Reve 2.1, and OpenAI's GPT Image 2, giving Microsoft the top two positions on that board. The model entered private preview on MAI Playground and Microsoft Foundry on August 19, per Microsoft's announcement — invite-gated, still no public API, still no price card.

The striking part is the trajectory, not just the rank. The entry that topped out at No. 2 on Arena's text-to-image board is literally named "mai-image-2.6-preview," carried roughly 3,500 votes in that snapshot, and climbed eight places with a +79 Elo gain over MAI-Image-2.5 in a single step. Since then the picture has filled in fast: private-preview access opened on MAI Playground and Microsoft Foundry on August 19, Arena's image-editing board moved the build up to No. 3, and now Artificial Analysis has it at No. 1 in image editing and No. 2 in text-to-image. That makes this the strongest independent signal yet that Microsoft's in-house image effort has closed the gap with OpenAI — and the public API still isn't generally available.

The key numbers

• Arena text-to-image rank: No. 2 — score 1,336±11, preliminary, ~3,488 votes

• Gap to No. 1 (GPT Image 2 Medium): 45 Elo

• Lead over No. 3 (Grok Imagine Image 2.0): 20 Elo

• Jump over MAI-Image-2.5: +79 Elo overall, +91 Elo on text rendering

• Predecessor's position: MAI-Image-2.5 at No. 10 (1,256)

• Image-editing rank: No. 3 on Arena's "Single Image Edit" — 1,420 pts, up from No. 5

• Artificial Analysis image-editing rank: No. 1 — ahead of MAI-Image-2.5-Pro (the previous No. 1), Reve 2.1, and GPT Image 2; Microsoft holds the top two spots

• Artificial Analysis text-to-image rank: No. 2 — behind only GPT Image 2; first in 5 of 19 category boards

• Private preview: announced Aug 19 on MAI Playground and Microsoft Foundry — vendor-stated, invite-gated, no price card yet

What MAI-Image-2.6 is

MAI-Image-2.6 is the newest model in Microsoft's in-house image generation line, built under Mustafa Suleyman's Microsoft AI. The family has moved fast: MAI-Image-1 was the foundation, MAI-Image-2 (March 2026) delivered the photorealism and text-rendering jump, MAI-Image-2.5 (June 2026) added professional-grade output and editing and became the engine inside Bing Image Creator, PowerPoint, and OneDrive, MAI-Image-2.5-Pro (July 2026) added an 8K premium tier — and 2.6 now lands roughly six weeks after its predecessor.

The cadence matters as much as the model. Microsoft is shipping image improvements on a six-to-eight-week cycle, closer to how frontier labs ship text models than to how the image market has historically moved. A No. 2 Arena debut at launch, and a No. 3 editing rank a week later, is what that cadence looks like once it compounds.

The leaderboard results, read carefully

Arena (the rebranded LMArena) ranks image models by blind human preference: voters see two images generated from the same prompt and pick the stronger one. Artificial Analysis runs the same kind of independent blind-preference arena with its own Elo ratings. Both are independent measures of what people actually prefer rather than vendor-run benchmarks — which is why a No. 2 debut carries real weight, and why two separate boards agreeing on the same model matters more than either number alone. It's also why the numbers need caveats: Arena's 1,336 score is preliminary, based on about 3,500 votes versus more than 70,000 for the model in first place, and the entry is labeled "preview." Early-vote Elo moves; treat the direction as solid and the exact score as provisional.

Where the gains are concentrated is more interesting than the headline number. Arena ranked MAI-Image-2.6 first in 3D imaging and modeling (up from sixth) and second in cartoon/anime/fantasy (up from eighth), product and branding and commercial design (up from seventh), text rendering (up from eighth, with the +91 Elo gain), and art (up from fourth). Microsoft says it improved in every measured category. That spread — commercial design, 3D, text, and art — is the signature of a general-purpose image model rather than a one-trick specialty.

Then two more data points arrived. Arena's image-editing board — the "Single Image Edit" category — lists the same mai-image-2.6-preview build at No. 3 with 1,420 points, up from No. 5 and 19 points ahead of MAI-Image-2.5, with text rendering the biggest single gain (+43 points). On August 25, Artificial Analysis went further: its Image Editing Leaderboard puts the preview build at No. 1, taking the top spot from Microsoft's own MAI-Image-2.5-Pro (previously No. 1 at 1,272 Elo) and sitting ahead of Reve 2.1 and OpenAI's GPT Image 2 — Microsoft now holds the top two editing positions on that board. Microsoft AI chief Mustafa Suleyman called 2.6 "the best image editing model on AA" and said Microsoft holds three of the top five positions — that's vendor framing; the underlying ranks are Artificial Analysis's own. The AA editing score for the preview build itself hasn't been published, and both boards are early and low-vote, so treat the direction as solid and the exact numbers as provisional.

What's actually new (vendor-reported)

On capabilities, Microsoft's own descriptions for MAI-Image-2.6 are:

• Stronger text rendering, portraits, and 3D imagery

• More polished commercial and photorealistic output across product, branding, and cinematic use cases

• Work across multiple reference images in a single generation

• Richer grounding — better adherence to the specifics in the prompt

• Greater control over reasoning, format, and resolution

Every one of those is vendor-reported and not yet independently reproduced. The independent evidence is the leaderboard pair: No. 2 in text-to-image on both Arena and Artificial Analysis, No. 3 on Arena's editing board, and No. 1 on Artificial Analysis' editing board — all preliminary. Until MAI-Image-2.6 is broadly available and third parties can run controlled evals, treat the capability list as Microsoft's claim and the two leaderboards as the working signal.

When you can actually use it

MAI-Image-2.6 went from "Arena only" to "Arena plus an official preview" this week. On August 19 Microsoft announced the model is available in private preview on MAI Playground and Microsoft Foundry. That's the vendor's own statement, and preview access is invite-gated rather than a public sign-up — so there is still no price card and no pay-per-token API. The practical tell to watch: when MAI-Image-2.5 became the default in Bing Image Creator and rolled into PowerPoint and OneDrive, it meant Microsoft considered it production-safe. A similar 2.6 rollout into those surfaces — or the "preview" label dropping — will be the signal that it's ready for production workloads.

What it might cost

No pricing for 2.6 has been published — the private preview has no rate card. The anchor is the rest of the family, all vendor-reported:

• MAI-Image-2.5: $5 per million text-input tokens, $8 image input, $47 image output

• MAI-Image-2.5-Flash: $1.75 / $1.75 / $19.50 — the high-volume tier

• MAI-Image-2.5-Pro (July 2026): $5 / $8 / $106 — 8K output, 96.8% text-rendering accuracy

That is a five-fold spread between the cheapest and priciest image-output rates Microsoft already sells, so 2.6's production cost will hinge on which tier it slots into. At 2.5's $47-per-million output rate, an image consuming roughly a thousand output tokens lands near five cents — but Microsoft does not publish per-image token counts, so treat that as arithmetic, not a quote.

Pricing is also where a routing layer earns its keep once the model reaches a public API. On OrcaRouter, whatever Microsoft charges is passed through at 0% markup — provider list price, nothing added — so a launch discount or a later price cut lands on your side of the bill the same day it's announced. No renegotiating and no second contract; the model shows up on the same key as everything else.

Where this leaves the image-model race

The top of Arena's text-to-image board right now: GPT Image 2 (Medium) at 1,381, MAI-Image-2.6 at 1,336, Grok Imagine Image 2.0 at 1,316, Reve 2.1 at 1,302, and Meta's Muse Image at 1,282. Five models within roughly 100 Elo at the top. Artificial Analysis reaches the same overall verdict — No. 2 in text-to-image, behind only GPT Image 2 — and adds category detail: the preview build takes first in 5 of its 19 text-to-image categories (Material, Knowledge, Frontier, Retail & Ecommerce, and Marketing & Advertising), with GPT Image 2 leading every other one. Two independent preference-based boards pointing the same direction is the strongest read you can get on a model this young. The quality gap that used to separate the leader from the pack has compressed to the point where "best image model" is no longer a useful question — editing, grounding, speed, and price are.

Microsoft's bet is broader than any single leaderboard crown: an in-house model stack spanning image, voice, code, and reasoning, distributed through Copilot, Bing, and Office. The Arena ranks are the proof-of-work; the distribution is the actual product.

How to try it without betting production on it

The Arena preview is free — go generate against it today if you want the raw signal. If you have private-preview access, the MAI Playground and Microsoft Foundry entry points let you run real workloads before you commit. For everyone else, the sensible way to adopt a brand-new model with a "preliminary" score is to route a slice of traffic to it and keep your proven model as the fallback. That is exactly the failure mode a routing layer is built for: on OrcaRouter you would point one endpoint at MAI-Image-2.6 once it is hosted on an upstream provider and let automatic failover catch the edge cases a low-vote leaderboard rating cannot see. One API key, 200+ models, provider prices passed through.

What to watch next

• Whether the Artificial Analysis No. 1 editing rank and the Arena No. 3 hold as votes accumulate — both boards are early and low-vote

• The public API price and which tier it lands in — the private preview still has no rate card

• Whether Bing Image Creator and Copilot switch their default from 2.5 to 2.6

• Independent benchmarks once the API opens

• Whether the "preview" label drops — the signal that Microsoft is confident enough to sell it

FAQ

When can I actually use MAI-Image-2.6?

Right now through Arena, which is free. Since August 19 it's also in private preview on MAI Playground and Microsoft Foundry — per Microsoft's own announcement, so access is invite-gated rather than a public sign-up. There is still no public, pay-per-token API and no published price.

How strong is the image-editing result?

On Artificial Analysis' Image Editing Leaderboard, MAI-Image-2.6-Preview is No. 1, ahead of MAI-Image-2.5-Pro, Reve 2.1, and GPT Image 2 — Microsoft holds the top two spots. On Arena's "Single Image Edit" board it's No. 3 at 1,420 points, up from No. 5. Both are independent blind-preference results but early and low-vote, and the AA editing score for the preview build hasn't been published.

How close is it to GPT Image 2 Medium?

45 Elo on the current text-to-image Arena snapshot — close enough that the No. 2 / No. 1 difference sits within what vote flow can move, and the top five models are all within roughly 100 Elo.

Are the Arena ranks reliable?

They are independent, blind human-preference results, which is exactly what makes them meaningful — but both are preliminary (about 3,500 votes on the text-to-image board) and attached to a preview build. Trust the direction, not the decimals.

Bottom line: MAI-Image-2.6-Preview is the first Microsoft image model that genuinely belongs in the frontier conversation — and it got there fast: No. 2 on Arena's text-to-image board at debut, No. 1 in image editing on Artificial Analysis by August 25, and private preview on Microsoft's own surfaces before any public API. Two independent boards now converge on the same reading. If the eventual price lands anywhere near the Flash tier, it becomes one of the most interesting value plays in image generation this year. The editing ranks and the preview answer "is this real." The public API and its price will answer "is this worth building on."