Microsoft MAI-Image-2.6 arrives on Vercel AI Gateway in standard and Flash tiers

Microsoft's MAI-Image-2.6 and MAI-Image-2.6-Flash are now live on Vercel AI Gateway. Developers can call either model with the model IDs microsoft/mai-image-2.6 and microsoft/mai-image-2.6-flash through the AI SDK, the AI CLI, or an OpenAI-compatible endpoint, using one gateway API key, unified billing, and request traces, with no separate Microsoft provider account.

That gives Vercel's user base access to the image models Microsoft introduced in September 2026, where the company ranked MAI-Image-2.6 No. 2 for text-to-image and image editing on Arena and No. 1 for image editing on Artificial Analysis. All three placements are vendor-reported, not independent lab results.

Standard tier or Flash: picking between the two

Vercel's changelog draws a clean line. Use microsoft/mai-image-2.6 for final assets, when image quality matters more than speed. Use microsoft/mai-image-2.6-flash for many iterations, prompt exploration, and variation batches, at lower cost and lower latency. Microsoft pitches Flash for latency-sensitive, high-throughput production workloads and, in an internal load test (100 requests per minute at 1024×1024), measured it 2.8x faster than GPT-Image-2-Medium with 72% greater efficiency. Those are Microsoft's own numbers, on Microsoft's chosen baseline, so treat them as vendor claims until third parties reproduce them.

Editing, reference images, and size limits

Both models generate images from text and edit existing ones. Each accepts up to five reference images, which you can use to guide an edit or combine subjects. Outputs reach 1536 × 1536 pixels. According to Microsoft's announcement, the family supports multi-image reference editing, web grounding, and dynamic aspect ratios, aimed at photoreal scenes, legible text rendering, and targeted changes to products, portraits, and layouts. A typical edit is a text instruction passed alongside the model, such as changing a vase to terracotta while keeping the flowers and setting intact.

Availability, cost, and what is missing

The models are closed: no open weights appear on Hugging Face, and the announcement lists Vercel AI Gateway and Microsoft Foundry as developer access points. Both IDs are already listed in Vercel's gateway model directory with per-model pages, though this post was published before checking those pages for exact per-image pricing. Confirm current rates in the gateway's model directory before wiring Flash into a high-volume pipeline. Availability is live now, not waitlisted, and a model playground is available for a no-code first pass.

What this changes for image-generation workflows

Teams already generating images through AI Gateway can swap Microsoft's models in with a model-ID change, keeping one bill and one key across providers. The more interesting question is placement. Microsoft positions MAI-Image-2.6 near the top of closed image models on cost-per-quality, and the Flash tier targets the variation-heavy work that agencies and commerce teams run dozens of times per asset.

Worth noting: Vercel's AI Gateway already carried OpenAI's Decisions lineup, and Google's Gemini 4 Argon reached GitHub Copilot the same way, model makers letting gateways handle distribution. A practical next step is to run the same brief through MAI-Image-2.6 and the incumbent model you pay for today, then compare edits side by side. LinkLoot's AI workflow automation guide and prompt library cover that kind of bake-off.