Microsoft moves its MAI image models out of preview in Foundry, with two price tiers

Microsoft’s image-generation models are moving beyond preview in Foundry. On October 27, 2026, the company made MAI-Image-2.6 and MAI-Image-2.6-Flash generally available, pairing an option aimed at detailed final assets with a cheaper model for repeated creative work. The release brings reference-guided editing, web context and format controls into that offering.

The change is the models’ availability status in Microsoft Foundry, not their first appearance. Vercel had already added both models to AI Gateway on October 13, offering generation and editing through its AI SDK or command-line tool with one API key and no provider account. Microsoft’s announcement now moves its own Foundry offering out of preview.

Five references instead of one long description

MAI-Image-2.6 accepts up to five reference images for editing. Microsoft describes this as a way to help preserve products, characters and brand assets while changing an image. A team can supply a product reference, a character and a visual direction together, rather than trying to describe every relevant detail from scratch in a text prompt.

The family also offers web grounding through Bing: generation can draw on relevant real-world context for prompts about recent or specific information. Microsoft explicitly says that this context is not a substitute for reviewing the output. Images still need review when factual accuracy, brand standards or editorial requirements matter.

Format controls address another part of the workflow: adapting an idea for different placements. Microsoft lists dynamic aspect ratios and output up to 1.5K resolution. Those controls sit alongside reference editing and web context; they let teams guide the shape and size of an asset as well as its content.

The cheaper tier has three separate rates

Microsoft publishes separate starting prices for text supplied to a model, images supplied to it and the image output it generates. The rates are quoted per million tokens, rather than as one flat charge per finished image. Flash is cheaper in all three categories, but the size of the difference varies by category.

Microsoft’s starting prices in US dollars

• Text input: $5 per million tokens for MAI-Image-2.6; $1.75 for Flash.
• Image input: $8 per million tokens for MAI-Image-2.6; $2.50 for Flash.
• Image output: $38 per million tokens for MAI-Image-2.6; $19 for Flash.

The clearest direct comparison is image output: Flash’s listed token rate is half the flagship’s. That is a comparison of published rates, not a promised price for an entire campaign or a fixed saving on every completed asset. Microsoft presents the two models as choices to evaluate against an application’s requirements, rather than naming one as the default for every task.

From variations to a final campaign image

Microsoft illustrates the division of work with a hypothetical retail campaign. A team uploads three references—a product photo, a brand color palette and a target aesthetic—then generates 10 background variations using web context for seasonal themes. It uses Flash for exploration and revisions before switching to MAI-Image-2.6 for a final main campaign image that needs text overlays.

That example is a suggested workflow, not a customer result. For actual testing, Microsoft points to Minute, the sports content and technology company that operates the Sports Illustrated publishing business. Minute says its early tests impressed it with visual quality, composition and targeted design edits, while describing broader creative uses as potential rather than a completed rollout.

Another distribution step remains ahead. Microsoft says MAI-Image-2.6 is coming to Adobe Firefly soon, and Adobe confirms that plan in the announcement. Neither gives a release date. Foundry’s general availability is the concrete change now; access inside Adobe’s creative AI studio is still a future integration.