Microsoft Opens MAI-Image-2.6 in Foundry Preview and Adds a Lower-Cost Flash Model

Microsoft has opened public-preview access to MAI-Image-2.6 through Microsoft Foundry and launched MAI-Image-2.6-Flash, a cheaper version aimed at applications that need fast, high-volume image work. The move gives developers a direct choice between the company’s flagship image quality and a model designed to reduce serving cost and response time.

Both models generate and edit images. They support multiple reference images, web grounding, and controls over aspect ratio, format, and resolution. Microsoft positions MAI-Image-2.6 as the precision option, while Flash is intended for latency-sensitive workloads, where a delay in returning an image or a high cost per request can shape the product experience.

One family, two operating targets

The flagship arrives with strong preference-benchmark positioning. Microsoft says MAI-Image-2.6 ranked No. 2 on Arena for both text-to-image generation and image editing on September 4. That is a quality signal, but it does not make Flash a like-for-like quality leader: Microsoft describes Flash as delivering comparable quality for production use rather than publishing the same ranking for the lower-cost model.

Speed claims use a different yardstick

Microsoft says Flash generates images 2.8 times faster than GPT-Image-2-Medium and delivers 72% greater efficiency. Those are company-supplied comparisons, not an independent measure of every production setup. Microsoft’s own launch material notes that latency can vary with the prompt and load, so the stated advantage is best read as a benchmarked performance claim rather than a guaranteed response time.

The price is metered by tokens, not pictures

The price gap is clearest on image output, but Foundry does not sell either model at a single fixed price per generated picture. It meters text inputs, image inputs, and image outputs separately. Image resolution, reference materials, and the output tokens a request consumes can all affect a developer’s invoice, making simple per-image estimates useful only under their stated assumptions.

• MAI-Image-2.6 starts at $5 per million text-input tokens, $8 per million image-input tokens, and $38 per million image-output tokens.
• Flash starts at $1.75 per million text-input tokens, $2.50 per million image-input tokens, and $19 per million image-output tokens.

Availability turns the score into a deployment decision

September 4 is the broader access point, not MAI-Image-2.6’s first appearance. Microsoft announced the model on August 10 and had placed it in its playground and a private Foundry preview by August 18. The public preview now lets developers test the more consequential questions on their own prompts: whether the flagship’s preference ranking holds for their brand rules and whether Flash keeps enough quality when speed and volume matter more.