Microsoft's MAI-Image-2.6 models arrive on Vercel AI Gateway with single-key access

What Vercel added

Vercel has brought Microsoft's MAI-Image-2.6 and MAI-Image-2.6-Flash image models to its AI Gateway, according to a changelog entry published on the company's blog on October 13. The addition means developers can generate images from text and edit existing pictures through Vercel's AI SDK or AI CLI, authenticating with a single API key and without opening a separate account with the model provider.

What the models do

According to Vercel, both variants support photorealistic output, text rendering, and targeted edits to products, portraits, and layouts — the kind of fine-grained manipulation that e-commerce and marketing workflows tend to need. Up to five reference images can be supplied to steer an edit or to combine subjects across images, and output sizes reach 1536 × 1536 pixels.

Text rendering is a notable inclusion in that capability list. Readable text inside generated images has historically been a weak point for image generators, so calling it out as a headline feature signals where Microsoft is aiming these models.

Two variants, two jobs

Vercel positions the pair as complementary:

• MAI-Image-2.6, addressed as microsoft/mai-image-2.6, is meant for final assets, where image quality matters more than generation speed.
• MAI-Image-2.6-Flash, addressed as microsoft/mai-image-2.6-flash, generates faster at lower cost, which Vercel frames as better suited to prompt exploration and producing many variations.

That quality-versus-cost split mirrors how image work tends to proceed in practice: quick, cheap iterations while exploring an idea, then a slower, higher-fidelity pass for the finished asset. Having both ends of that spectrum behind the same API makes it easy to switch between them as a project moves from drafts to deliverables.

How to use it

Both models are exposed through the same interface as other models on the gateway. According to the changelog, image generation is a call to the SDK's generateImage function with a text prompt, while editing is done by passing an existing image as input together with a natural-language instruction. Vercel also points developers to its model playground for trying the models before writing any code, and to its image generation guide for integration details.

Why it matters

The significance here is mostly about access and plumbing rather than new model capabilities. Teams already building on Vercel's AI SDK can now route image generation and editing to Microsoft's models alongside whatever else they use, with one key and one integration surface — no separate provider signup and no second authentication path to maintain. For Microsoft, appearing in a third-party gateway widens distribution of its image models to developers who might otherwise never provision direct access.

The two-tier lineup also matters for cost discipline. Iteration-heavy image work — generating variations, testing prompts, mocking up layouts — burns through API calls quickly, and a cheaper, faster Flash variant gives teams a way to keep exploration inexpensive while reserving the full-quality model for final output. For anyone already shipping AI features on Vercel, this is a small integration change that removes a genuine piece of operational overhead.