Summary: Microsoft AI introduced MAI Image 2.7 and MAI Image 2.7 Flash on November 17, 2026, alongside Canvas, a new interactive image workspace built on the Flash model. Both models are in public preview on Microsoft Foundry today, priced per finished image: roughly $35 per 1,000 images for the flagship and $10 per 1,000 for Flash. If you already pay leading-competitor prices for image generation, the price-to-quality claim here is worth testing in your own pipeline — but the evidence is Microsoft’s own, so test rather than switch.
What actually shipped
Three things, in one announcement. First, MAI Image 2.7, the flagship model for image generation and editing. Second, MAI Image 2.7 Flash, a faster variant that Microsoft positions for production workloads. Third, Canvas, an interactive workspace powered by Flash where users create, edit, and refine images — a product surface, not just an API endpoint.
The availability story matters for adoption: this is a public preview on Microsoft Foundry, not a waitlist. A builder can wire it up today. That is not the first time MAI models have reached Foundry as previews; for context on how Microsoft has staged these, see our earlier note on Microsoft MAI models in Foundry preview.
The price math, per finished image
Pricing is quoted per output, which is the unit that matters for a finished task: ~$35 per 1,000 images for MAI Image 2.7, ~$10 per 1,000 for Flash. At the flagship price, one image costs about 3.5 cents; Flash is about a cent.
Microsoft’s cost claim is sharper than the list price. The company states that MAI Image 2.7 takes the #1 spot on Artificial Analysis image editing, outperforming GPT Image 2.6 Sunburst (max) at one-fifth of the cost. Note what that sentence actually ties together: the one-fifth figure is attached to that specific editing comparison, not to every workload or every competitor. Microsoft also says both models sit on the quality-cost Pareto frontier for generation and editing.
Where the quality claims come from
Every performance number in the announcement is a leaderboard citation: the Artificial Analysis editing rank and a top-two lab position on Arena for both image generation and editing, where Microsoft says MAI outperforms GPT Image 2.6 Flare. These are results Microsoft cites in its own post — this publication has not run the models, and the announcement gives no independent verification. A vendor’s own benchmark is a reason to run a test, not a reason to switch.
Two things the post does not state would also decide adoption for some teams: rate limits and data terms for the preview. Retention and training-use policies are product terms, not fine print — they decide whether a tool can be adopted at all, especially for teams handling customer images. The announcement page goes further than we can quote here, so check the full terms on Foundry before committing anything sensitive.
Our take: Run a side-by-side bake-off this week — but treat it as an experiment, not a migration. The case for acting now: the models are in a usable public preview, pricing is per finished image with no token accounting to model, and the flagship costs about $35 per 1,000 images. If your current tool charges five times that for comparable quality — Microsoft’s claimed editing comparison — then even if MAI Image 2.7 only matches your current output, an afternoon of testing against your real prompts costs almost nothing and reverses easily. We would especially push small teams already paying leading-competitor prices at volume to try Flash first: at ~$10 per 1,000 images, it sets the floor. Teams whose current tool already clears their quality bar, and whose image volume is low enough that a fourfold price gap saves pocket change, can ignore this release. What would change our position: your own outputs falling short of the claimed quality, or Foundry preview terms that block your data requirements — either one, and the current tool stays.
Sources
View the original post on X
