MAI-Image-2.5-Pro vs MAI-Image-2.6 Preview: The Editor You Can Deploy vs the Model You Can Only Rank

The question everyone is asking about MAI-Image-2.5-Pro and MAI-Image-2.6 Preview is which one is better. That is the wrong question, because only one of them is a thing you can build on. MAI-Image-2.6 Preview is the model that ranks higher on the text-to-image leaderboard, but MAI-Image-2.5-Pro is the one that has a public API, a published price card, and the top-ranked image editor on an independent board. Microsoft shipped both within a month of each other, and the reason the pairing confuses people is that the two occupy different points in the same product pipeline — one is a product you can deploy today, the other is a ranked preview of what the family becomes next.

The one-sentence version

If you need a Microsoft image model in production now, the only choice is MAI-Image-2.5-Pro — public preview, token-metered, #1 on the independent image-editing leaderboard. If you are choosing what to build against for the next quarter, MAI-Image-2.6 Preview is the direction of travel: it outranks its sibling at plain text-to-image, but it is invite-gated, unpriced, and not yet a callable API. One is a deployment; the other is a signal.

Two releases, a month apart

Microsoft announced MAI-Image-2.5-Pro on 23 July 2026, alongside MAI-Voice-2-Flash, and placed it in public preview on Azure AI Foundry across seven global regions plus the free MAI Playground. It is the quality tier of the in-house MAI-Image-2.5 family — the same family that already powers Bing Image Creator, PowerPoint, and OneDrive — and it carries the family's highest-fidelity positioning: hero imagery, detailed editing, and precise in-image text.

MAI-Image-2.6 Preview is the next step. The build surfaced on Arena's text-to-image leaderboard around 10 August, and on 19 August Microsoft confirmed it in private preview on MAI Playground and Microsoft Foundry. So within four weeks the company shipped its premium tier and then previewed its successor — a cadence that matters for this comparison more than any single number.

The two boards disagree, and that is the whole story

Part of the confusion is that the two models lead on different leaderboards, and neither has topped the same one.

• MAI-Image-2.5-Pro — No. 1 on the Artificial Analysis Image Editing Arena at Elo 1,272 (roughly 6,100 blind votes), ahead of Reve 2.1 at 1,262 and MAI-Image-2.5 and GPT Image 2 (high) tied at 1,256. On Artificial Analysis's Text-to-Image board it is a more modest No. 7 at Elo 1,292.

• MAI-Image-2.6 Preview — No. 2 on Arena's text-to-image board at Elo 1,336 (~3,500 votes), behind GPT Image 2 Medium at 1,381 and ahead of Grok Imagine Image 2.0 at 1,316. Arena's image-editing board lists the same preview build at No. 3 with 1,420 points, up from No. 5.

The boards are different boards — Artificial Analysis versus Arena — so the numbers are not directly comparable to each other, and every figure here is a low-vote preliminary. But read them for direction and the picture is consistent: MAI-Image-2.6 Preview is the better generator from a blank prompt, and MAI-Image-2.5-Pro is the better editor of an existing image. Microsoft has not shipped a single model that tops both, and the two current leaders have chosen different halves of the job.

What you can actually call today

This is where the comparison stops being academic.

• MAI-Image-2.5-Pro — deployable. Public preview on Azure AI Foundry in seven regions, free on MAI Playground, token-metered with a rate card. You can sign up, send an image and an instruction, and get a result back today.

• MAI-Image-2.6 Preview — invite-gated. Private preview means access is granted, not purchased, and there is no price card and no pay-per-token endpoint. The Arena build is free to vote against but is not a service you can call from code.

If your requirement is a live API, the decision has already been made for you. MAI-Image-2.6 Preview is a reason to wait, not a thing to adopt.

The Pro rate card, and what it buys

MAI-Image-2.5-Pro is token-metered on Foundry: $5 per million text input tokens, $8 per million image input tokens, $106 per million image output tokens. Microsoft does not publish tokens-per-image, so per-image cost is arithmetic rather than a quote; using Artificial Analysis's conversion, a 1024×1024 image lands near $108.50 per 1,000 images. That is roughly 2.3× the sibling MAI-Image-2.5 (~$48 per 1k) and 5.4× MAI-Image-2.5-Flash (~$20 per 1k) — the Pro tier is a deliberate premium for the editing crown.

The documented spec sheet for the Pro tier: up to 32,000 text input tokens, a 131k context window, 4,096 output tokens, JPEG/PNG image input, and output capped at 1,048,576 pixels (a 1024×1024-equivalent). That last number is worth flagging because Microsoft's marketing touts an "8K" tier; on the API, the practical ceiling is about a megapixel. For production buyers the working resolution is ~1K, not print resolution.

MAI-Image-2.6 Preview has no rate card yet. The anchor for what it might cost is the family's existing spread — $1.75 to $5 input, $19.50 to $106 output across Flash, 2.5, and Pro — so 2.6's production price will hinge on which tier it slots into, and preview pricing is not GA pricing regardless.

What MAI-Image-2.6 Preview adds when it lands

Microsoft's own description of the 2.6 upgrade, all vendor-reported and none independently reproduced yet:

• Stronger text rendering — longer sentences, complex layouts, and the single biggest Elo gain over 2.5 (+91 on Arena's text-rendering category, per Arena's published breakdown)br />• More natural portraits — facial detail, lighting, skin texturebr />• Better 3D imagery and modeling, plus stronger grounding — closer adherence to the specifics in the promptbr />• Multi-reference-image support in a single generation, and more user control over reasoning, format, and resolution

Notice what that list does not include: nothing about beating the Pro tier at editing. Microsoft is describing 2.6 as a better generalist, not as a better editor than its own premium tier. That is why the two models can coexist in the lineup — the editing specialist and the fast-improving generalist are different products.

The sequencing is the strategy

Microsoft is shipping image improvements on a six-to-eight-week cycle, and the Pro-vs-preview pairing is what that cadence looks like from the outside. The pattern across the family has been consistent: a model enters preview, rolls into Bing Image Creator and Office surfaces once it is considered production-safe, and the next model starts its preview cycle. MAI-Image-2.5 already runs 100% of Bing Image Creator's generation and is the engine in PowerPoint and OneDrive; the Pro tier is the same family pointed at higher-fidelity work. MAI-Image-2.6 Preview is where that pipeline is heading next.

The practical read: if you are building on the family, adopt MAI-Image-2.5-Pro now for editing workloads and expect MAI-Image-2.6 (or a Pro successor) to arrive on the same surfaces within a quarter. The independent leaderboard gives the Pro tier a defensible claim — the best-regarded image editor on a blind-vote board — that no vendor announcement can manufacture.

Where a routing layer fits — eventually

Neither model is callable through a third-party API yet, so neither is reachable on OrcaRouter — and an honest comparison should say so plainly. When MAI-Image-2.5-Pro (or, later, MAI-Image-2.6 Preview) becomes available through a callable hosted provider, the pass-through economics apply: OrcaRouter charges 0% markup over the provider's list price, so Microsoft's rate card is what you pay, and a launch discount or price cut lands on your bill the same day it is announced — no renegotiation, no second contract. The other half of the argument for a routing layer is failover: a brand-new preview model with a low-vote leaderboard score is exactly the thing you route a slice of traffic to while keeping a proven model as the fallback, rather than betting a production path on an unpriced preview.

Who should act, and who should wait

Deploy MAI-Image-2.5-Pro now if your workload is changing existing images — targeted edits, in-image text, product hero work — because it is the only one of the two with an API, and its #1 editing rank is independent evidence it does the job. Wait on MAI-Image-2.6 Preview if you are generating from a blank prompt and quality is the only consideration: the preview build already outranks its sibling at text-to-image, and its eventual Foundry price will likely undercut the Pro tier. And if you are choosing a family to standardize on, treat 2.6 Preview as the signal of where Microsoft's image line is heading — not as a model you can use today.