Microsoft releases MAI Image 2.7, with Flash priced around $10 per 1,000 images
The image-generation and editing release adds transparent backgrounds, larger outputs and a Flash-powered workspace, with developer access beyond Microsoft Foundry.
Loading page…
The image-generation and editing release adds transparent backgrounds, larger outputs and a Flash-powered workspace, with developer access beyond Microsoft Foundry.
Listen to this story
For teams choosing between quality and iteration cost, Microsoft is positioning MAI Image 2.7 for final assets and Flash for faster, cheaper exploration. Both support outputs up to 2048 × 2048 pixels, transparent backgrounds, and editing guided by as many as five reference images. Vercel provides access through its AI Gateway without a Microsoft account, while OpenRouter’s full-model pricing is split across input and output tokens; its listing does not currently offer EU in-region routing.
Microsoft says the full model ranks first on Artificial Analysis image editing, ahead of GPT Image 2.6 Sunburst (max), at one-fifth its cost; it gives no separate ranking for Flash.
Vercel recommends the full model for quality-focused final assets and Flash for faster, lower-cost prompt exploration and variations.
Vercel supports both models through its AI Gateway and CLI; its SDK integration requires AI SDK 6 or later.
Microsoft is pitching its latest image models as a way to buy leading editing quality without leading-model prices. On November 17, 2026, it released MAI Image 2.7 and MAI Image 2.7 Flash in public preview on Microsoft Foundry, listing approximate costs of $35 and $10 per 1,000 images, respectively.
Alongside the models, Microsoft launched Canvas, an interactive workspace for creating, editing and refining images. It runs on Flash, putting the lower-cost model behind a visual workspace as well as making it available to developers.
Vercel, which added both models to its AI Gateway, describes the distinction in practical terms: use the full model for final assets where quality takes priority, and Flash for faster, cheaper prompt exploration and variations.
Microsoft says MAI Image 2.7 ranks first on Artificial Analysis image editing, ahead of GPT Image 2.6 Sunburst (max), at one-fifth of its cost. It also places MAI among the top two labs on Arena for image generation and editing, outperforming GPT Image 2.6 Flare. Those claims do not establish a separate ranking for Flash.
Microsoft’s approximate cost for the full model in public preview.
Microsoft’s approximate cost for the Flash model in public preview.
OpenRouter’s listing for the full model separates charges into $5 per million text-input tokens, $8 per million image-input tokens and $34 per million image-output tokens. Those are token rates, not a flat per-image tariff: the listing charges separately for text supplied, images supplied and images generated.
Vercel’s launch changelog says both models increase the maximum output size to 2048 × 2048 pixels and add transparent backgrounds for generated and edited images. It lists photorealistic generation, text rendering and targeted edits to products, portraits and layouts.
Vercel illustrates the generation-to-editing workflow with a perfume bottle: first request it against a transparent background, then pass the bottle image to Flash with instructions to place it on a marble counter.
Vercel offers both models through one AI Gateway API key without a Microsoft account. Its integration supports AI SDK 6 or later, with either model selected by its identifier. Generation and editing are also available through Vercel’s command-line tool using the same key and model identifiers.
On OpenRouter, the full model is served through Azure. For developers who need EU in-region processing, that route has a specific limit: OpenRouter says EU in-region routing is not currently available for this model.
Loading discussion...
Join the conversation
Explain when more options help you make a better choice.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.