Summary: On October 13, 2026, Vercel announced that Microsoft’s MAI-Image-2.6 and MAI-Image-2.6-Flash are available through AI Gateway. You can generate or edit images through the AI SDK or AI CLI with one API key and no separate provider account. The practical decision is simple: the standard model for final assets, Flash for cheap, fast iteration.
The tradeoff the two models encode
Vercel positions the pair explicitly. MAI-Image-2.6 (microsoft/mai-image-2.6) is for final assets, when image quality matters more than speed. MAI-Image-2.6-Flash (microsoft/mai-image-2.6-flash) generates faster at lower cost, aimed at the exploration phase — trying prompts, producing variations.
That split maps cleanly onto how image features actually get built. During development you burn through dozens of generations refining a prompt; paying premium rates for throwaway drafts hurts. Once the prompt is locked, you swap the model ID and regenerate the shipped asset at full quality. Because both models sit behind the same generateImage call, that swap is a one-string change, not a refactor.
What the models can do
Per Vercel’s changelog, both models support photorealistic imagery, text rendering, and targeted edits to products, portraits, and layouts. Two capabilities stand out for product work:
- Reference images. You can pass up to five reference images to guide edits or combine subjects. That’s the difference between “make the hero shot blue” and hoping the model guesses right.
- Output up to 1536×1536. Large enough for most marketing surfaces without a separate upscaling step.
The wiring is three lines
Generation goes through the AI SDK’s generateImage:
import { writeFile } from 'node:fs/promises';
import { generateImage } from 'ai';
const { image } = await generateImage({
model: 'microsoft/mai-image-2.6',
prompt: 'A blue ceramic vase with wildflowers on a sunlit wooden table.',
size: '1024x1024',
});
await writeFile('image.png', image.uint8Array);Editing is the same call with images attached to the prompt:
const { image: editedImage } = await generateImage({
model: 'microsoft/mai-image-2.6',
prompt: {
text: 'Change the vase to terracotta and keep the flowers and setting.',
images: [image.uint8Array],
},
});Vercel also lists an OpenAI-compatible path and an AI CLI option, so you’re not locked into the SDK if your stack already speaks that protocol.
Why the gateway matters more than the models
One key, no provider account — that’s the real shipping story here. If you’re already routing models through Vercel’s AI Gateway, adding an image model is a config change, not a procurement task. That pattern compounds: each new model drops into the same billing, same observability, same code path, which keeps the marginal cost of trying something new near zero.
Vercel’s changelog does not specify pricing for either model, so check the model details page before committing Flash to a high-volume pipeline — “lower cost” is directional, not a number.
Practical next step
If you have an image feature on the roadmap, the fastest way to evaluate is the model playground Vercel links from the announcement. Test your real prompts on both model IDs; if Flash output is good enough for drafts and the full model only runs at export time, you’ve structured the feature to pay for quality exactly once.
Sources
View the original post on X