Summary:
Flash adds a lower-priced option for image generation and editing. Hosted API access, token-based charges and the Canvas preview have different boundaries.Microsoft's November 17 rollout adds MAI Image 2.7 Flash, the Canvas workspace and broader API access to its image-model lineup. The new story is not simply another editing-preview announcement: it gives developers a lower-priced variant and a clearer route to generation and editing through hosted services.
Our October coverage of the original MAI Image 2.7 preview described a narrower editing release. The current Microsoft announcement places both 2.7 and Flash in public preview on Foundry, while Canvas uses Flash for interactive creation and refinement.
What developers can use now
Vercel confirms public-preview availability for both models through AI Gateway. Its announcement documents text-to-image generation, edits guided by reference images, output up to 2048 × 2048 pixels and transparent backgrounds. Developers can use a Gateway key without a separate Microsoft account.
The model IDs are microsoft/mai-image-2.7 and microsoft/mai-image-2.7-flash. Vercel positions the larger model for final assets and Flash for quicker exploration and variations. The announcement's examples use AI SDK 6 or later; transparent output is requested through the Azure provider's background option.
Reference-image limits depend on the endpoint. Vercel's launch documentation describes up to five references, while fal documents up to two for its editing endpoints. Check the chosen API's schema rather than treating one provider's limit as universal.
Compare token rates, not a flat image price
The Gateway listing for MAI Image 2.7 shows US$5 input and US$34 output per million tokens. The Flash listing shows US$1.75 input and US$10 output per million tokens. These are listed rates checked November 18, not a fixed quote for every generated image.
| Listed rate, USD per million tokens | MAI Image 2.7 | MAI Image 2.7 Flash |
|---|---|---|
| Input | $5 | $1.75 |
| Output | $34 | $10 |
Bold identifies the lower listed price, not better image quality. Input and output are separate charges. Fal illustrates the distinction with a 1024 × 1024 image: 1,024 output image tokens cost about $0.035 on 2.7 or $0.010 on Flash, plus input tokens. Higher-resolution output and repeated attempts can change the bill.
Canvas is a separate access path
Microsoft describes Canvas as an interactive workspace for turning prompts or sketches into images and selecting areas to replace, move or restyle. The current Canvas page exposes prompt and diagram modes alongside image-editing controls.
Do not confuse that consumer preview with a commercial API entitlement. The MAI Labs entry page says Labs is for personal, non-commercial use and excludes school or work activities. Fal separately says content generated through its API can be used commercially, subject to its terms. Choose the service appropriate to the project before uploading client material.
A practical way to evaluate Flash
Start with a small, repeatable set of tasks: a clean product cutout, a localized color change and an image containing short text. Count accepted outputs rather than just generations. Record the input and output charges, retry count and time spent correcting lettering, edges or unintended changes.
That makes a lower-priced model useful where it actually reduces the cost of a finished asset. Keep the larger model as a comparison for difficult edits, and evaluate the two on the same reference material and acceptance criteria.
Sources and image credit
Sources: Microsoft's November 17 announcement and MAI Labs access notice; Vercel's launch documentation and model listings; fal's model guide, all linked above. Cover: original PromptsZone chart of Vercel's listed output-token rates. The official Microsoft provider logo is used under Microsoft News Center's editorial-use permission; it identifies the provider, not a separate MAI model logo.


