Microsoft's MAI-Image-2.6 and Flash Models Arrive on Vercel AI Gateway
Vercel has added Microsoft's MAI-Image-2.6 and MAI-Image-2.6-Flash to its AI Gateway. Developers can generate images from text or edit existing ones through the AI SDK or the AI CLI, using one API key and no separate provider account.
- Microsoft's MAI-Image-2.6 and MAI-Image-2.6-Flash are now available on Vercel's AI Gateway.
- Both models support text-to-image generation and editing, including photorealism, text rendering and targeted edits.
- Up to five reference images can guide edits or combine subjects, with output sizes up to 1536 × 1536.
- MAI-Image-2.6 targets final assets, while the Flash version is faster and cheaper for iteration, according to Vercel.
- Access requires one API key and no provider account, via the AI SDK or AI CLI.
Two models for different workflows
According to Vercel, both models handle photorealistic imagery, text rendering, and targeted edits to products, portraits, and layouts. Users can supply up to five reference images to guide an edit or to combine subjects, and output sizes reach 1536 × 1536.
The two models are positioned for different needs:
- MAI-Image-2.6 (microsoft/mai-image-2.6) is meant for final assets, when image quality matters more than speed.
- MAI-Image-2.6-Flash (microsoft/mai-image-2.6-flash) offers faster generation at a lower cost. Vercel recommends it for cases that need many iterations, such as exploring prompts or generating variations.
In practice, this suggests a split where teams draft and experiment with the Flash model, then switch to the standard model for the output they intend to ship.
How to use them
Access works through the model ID, which is either microsoft/mai-image-2.6 or microsoft/mai-image-2.6-flash. Vercel's example for text-to-image calls the generateImage function from the ai package with a model ID, a text prompt and a size. The sample prompt describes a blue ceramic vase with wildflowers on a sunlit wooden table, requested at 1024x1024. The returned image is then written to a file as a PNG.
Editing follows the same pattern. Instead of a plain string, the prompt becomes an object containing a text instruction and an array of input images. In Vercel's example, the earlier image is passed back with the instruction to change the vase to terracotta while keeping the flowers and setting. The result is saved as a new file.
Vercel's changelog shows code for the AI SDK, an OpenAI-compatible option and the AI CLI, so developers can pick the interface that fits their existing setup.
What it means for developers
The main change is access. Through AI Gateway, teams can use Microsoft's image models without opening a provider account of their own, and with a single API key. For those already routing other models through the gateway, image generation and editing can sit alongside them in the same setup.
Vercel points developers to its model playground to try MAI Image 2.6 and MAI Image 2.6 Flash, and to its image generation guide for further detail. The changelog does not give pricing or performance figures for either model.
