MAI-Image-2.6 and MAI-Image-2.6-Flash: Quality and speed at production scale

Image generation has progressed significantly, leaving the experimental phase behind. Today, marketing teams are producing campaign materials, retailers are assembling their catalogs, and product teams are generating images directly within the tools they already utilise. This evolution shifts the focus: a model must now do far more than just create a visually appealing image. It should be capable of generating clear text, ensuring consistency across various formats, and accurately following creative directives.

We are excited to present MAI-Image-2.6 and MAI-Image-2.6-Flash in Microsoft Foundry, the latest advancements in the Microsoft AI image generation suite.

MAI-Image-2.6 offers a significant enhancement in quality when compared to MAI-Image-2.5, introducing features such as multi-reference editing, web grounding, and improved control over format and resolution. Meanwhile, MAI-Image-2.6-Flash is designed for production tasks that demand speed and efficiency, outperforming GPT-Image-2-Medium by over 2 times and being 78% more efficient than GPT-Image-2.

With MAI-Image-2.6, you’ll find strong external benchmark results across both Arena leaderboards:

• 2nd position on the Arena text-to-image leaderboard at launch.

• 3rd position on Arena’s image-editing leaderboard, ahead of Google’s Nano Banana family and Meta’s Muse Image.

Bar chart showcasing MAI-Image-2 among other text-to-image models on Arena.ai rankings

• Multi-reference editing. MAI-Image-2.6 allows you to include up to five reference images in a single request. This means that a locked creative element—be it a product, character, or brand logo—remains consistent as you iterate and resize it. Coupled with adaptive aspect ratios and outputs in 1.5K resolution, one approved concept can easily adapt to various formats without losing fidelity.
• Web grounding. This model can retrieve real-world context rather than solely relying on its training data, resulting in images that include accurate depictions of places, objects, and subjects.
• Enhanced control over format and resolution. Developers have precise controls over how thorough the model is during generation, the shape of the output, and the detail level—allowing it to cater to both rapid iteration cycles and high-quality final outputs.

Multi-reference feature: merge your brief, brand assets, and up to 5 references into a cohesive direction

MAI-Image-2.6-Flash is specifically optimised for situations where generating numerous variations quickly is critical, or for handling content at scale.

It is over 2 times faster than GPT-Image-2-Medium and 72% more efficient than GPT-Image-2. This efficiency makes it ideal for interactive applications, automated workflows, personalisation efforts, and high-volume creative production.

Select MAI-Image-2.6 when top-notch precision and output quality are paramount, such as for final campaign materials, complex editing, intricate commercial designs, and high-fidelity text production.

Opt for MAI-Image-2.6-Flash when speed, throughput, and efficiency are your primary concerns, including interactive experiences, fast iterations, personalisation, and bulk generation.

Teams can leverage Flash to brainstorm concepts and variations quickly, before switching to MAI-Image-2.6 for final assets that demand higher precision.

MAI-Image-2.6 and MAI-Image-2.6-Flash are now available in public preview in Foundry. The pricing for MAI-Image-2.6 starts from $5 USD per million tokens for text input, $8 USD per million tokens for image input, and $38 USD per million tokens for image output. Meanwhile, MAI-Image-2.6-Flash pricing begins at $1.75 USD per million tokens for text input, $2.50 USD per million tokens for image input, and $19 USD per million tokens for image output.

Frequently Asked Questions

• What is MAI-Image-2.6? MAI-Image-2.6 is a new image generation model by Microsoft that enhances image quality and allows greater control over image attributes.
• How does MAI-Image-2.6 differ from MAI-Image-2.5? MAI-Image-2.6 offers improved quality, additional features like multi-reference editing, and better control over output format and resolution.
• When should I use MAI-Image-2.6 versus MAI-Image-2.6-Flash? Use MAI-Image-2.6 for high-quality final outputs and complex edits, while MAI-Image-2.6-Flash is best for quick iterations and high-volume generation.
• Where can I access MAI-Image-2.6 and MAI-Image-2.6-Flash? Both models are available for public preview in Microsoft Foundry.
• What are the pricing details? Pricing for MAI-Image-2.6 starts from $5 USD per million tokens for text inputs, while MAI-Image-2.6-Flash starts from $1.75 USD per million tokens.