MAI-Image-2.7 API and Pricing, Explained Through a Real Pipeline
MAI-Image-2.7 is Microsoft AI's newest image generation and editing family, in a precision tier and a Flash tier that take the same inputs. Both are priced per output image token, which works out to about 3.5 cents an image on MAI-Image-2.7 and about 1 cent on Flash. Here is the rate card, what our probes rendered and how long they took, what a reference costs, how it compares with MAI-Image-2.6, and how to reach either tier through a published API endpoint.

MAI-Image-2.7 is Microsoft AI's newest image generation and editing family. It reached OpenRouter on November 17, 2026, and joined the Image Generation node, chat and its own model page the next day, after we probed both tiers through the same path the node uses. Like the generation before it, it comes in two tiers with one spec sheet: MAI-Image-2.7 for design-ready stills and MAI-Image-2.7 Flash for speed and volume. Both generate from a prompt, and both edit or compose from a prompt plus up to five reference images.
The headline is the price. The precision tier lists a little below MAI-Image-2.6, and the Flash tier lists at about half of 2.6 Flash, which puts it near a cent an image. This post covers the rate card, how it becomes a flat per-image price, what our test renders looked like and how long they took, what a reference adds, and how to call either tier as a published API endpoint.
The spec sheet
| Capability | MAI-Image-2.7 | MAI-Image-2.7 Flash |
|---|---|---|
| Input | Text and image | Text and image |
| Reference images | Up to 5 per request | Up to 5 per request |
| Aspect ratios | 7, plus auto | 7, plus auto |
| Images per request | 1 | 1 |
| Output at 1:1 | 1024 x 1024 | 1024 x 1024 |
| In chat | Yes | Yes |
Every row matches. The seven ratios are 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3, so one node covers a square thumbnail, a vertical story frame and a wide header, and auto lets the model choose from the prompt. There is no resolution setting on either tier. A pipeline built on one tier moves to the other by changing the model on the node and nothing else.
The provider also lists a transparent background option for this generation. We have not tested it yet, so plan on an opaque image, and use the Remove Background node when you need a cut-out today.
What MAI-Image-2.7 costs per image
These are the listed prices as of November 2026. Both tiers are priced per token:
| Tier | Per output image token | Per input image token | Per 1024 token image |
|---|---|---|---|
| MAI-Image-2.7 | $0.000034 | $0.000008 | about $0.035 |
| MAI-Image-2.7 Flash | $0.00001 | $0.0000025 | about $0.010 |
The last column is the one to plan with. Every image in our probes came back as exactly 1024 output image tokens, so the token rate turns into a flat price: about 3.5 cents an image on MAI-Image-2.7 and about 1 cent on Flash. The precision tier costs about three and a half times as much as Flash, a wider gap than the two-to-one split in the generation before it.
A hundred Flash images list at about $1.02. The same hundred on MAI-Image-2.7 list at about $3.48.
What a reference image adds
References bill as input image tokens. Our test reference, a 1280 x 853 product photo, read as 1,659 input image tokens. At the listed rates that is about 1.3 cents on MAI-Image-2.7 and about 0.4 cents on Flash, so a one-reference edit lists at about 4.8 cents and about 1.4 cents. A smaller reference reads as fewer tokens and costs less. The text of the prompt itself adds a fraction of a cent.
Against MAI-Image-2.6
| MAI-Image-2.6 | MAI-Image-2.7 | 2.6 Flash | 2.7 Flash | |
|---|---|---|---|---|
| Per output image token | $0.000038 | $0.000034 | $0.000019 | $0.00001 |
| Per 1024 token image | about $0.039 | about $0.035 | about $0.0195 | about $0.010 |
The precision tier is about 10 percent cheaper than the one it follows. Flash is the bigger change: just under half the price of 2.6 Flash for the same inputs. Both generations stay in the picker, so a pipeline already tuned on 2.6 keeps working until you decide to switch.
What we saw on November 18
We probed both tiers on November 18, 2026, at 1:1 through the chat completions route the node uses. Each tier got the same text prompt and the same edit:
| Probe | Tier | Wall time |
|---|---|---|
| Text to image | MAI-Image-2.7 | about 29 seconds |
| Text to image | Flash | about 11 seconds |
| Edit with one reference | MAI-Image-2.7 | about 17 seconds |
| Edit with one reference | Flash | about 12 seconds |
The text prompt asked for a ceramic cup of coffee on a sunlit wooden cafe table, steam rising, shallow depth of field, natural light. Both tiers returned a speckled stoneware cup with latte art on a worn wood table, with books, a glass vase and a chalkboard behind. MAI-Image-2.7 spent its extra time on surface detail in the glaze and the grain, and the composition reads as a styled product shot. It also added lettering nobody asked for, on book spines, a notebook and the chalkboard, and the first word on the chalkboard came out garbled. Flash put clean, readable words on its chalkboard and books.
So neither tier is a text renderer you can count on. If an image needs words, keep them out of the prompt and add them afterwards with the Overlay or Captions nodes, where the type is real.
The edit gave both tiers a still of a folded navy linen shirt and asked for the same shirt on a white marble counter in a bright studio. Both kept the shirt: the colour, the linen weave, the collar, the dark buttons and the fold. Both set it on veined white marble in soft daylight with a sprig of greenery at the back. MAI-Image-2.7 came in a touch closer on the weave; Flash pulled back a little to show the counter's edge. Either one is usable as a product still.
Two renders per tier on one day is not a benchmark, and we make no claim beyond what those four images show. The pattern was consistent, though: Flash in 11 to 12 seconds, the precision tier in 17 to 29.
Flash to draft, MAI-Image-2.7 to finish
With two tiers on one spec sheet and a three-and-a-half-to-one price ratio, the working pattern is simple:
- Draft on Flash. Try framings and prompt wording at about a cent a frame. Twenty-five variations list at about 26 cents.
- Finish the keeper on MAI-Image-2.7. Rerun the chosen prompt once on the precision tier for about 3.5 cents. The prompt, the ratio and the references carry across unchanged.
- Keep the product in every draft. At about 0.4 cents a reference on Flash, there is no reason to draft without it.
On the node that is one change in the model menu. In chat, ask for MAI-Image-2.7 Flash by name while you explore, then ask for MAI-Image-2.7 on the frame you keep.
Where it sits in the price band
Listed prices per image as of November 2026 for the image models around it:
| Model | Per image | Notes |
|---|---|---|
| MAI-Image-2.6 | about $0.039 | up to 5 references |
| Seedream 5.5 | $0.035 | 2K only, $0.008 per reference, up to 10 |
| MAI-Image-2.7 | about $0.035 | up to 5 references |
| MAI-Image-2.6 Flash | about $0.0195 | up to 5 references |
| Seedream 5.0 Flash | $0.018 | |
| MAI-Image-2.7 Flash | about $0.010 | up to 5 references |
| FLUX 3 Image Klein | $0.009 at 1K |
MAI-Image-2.7 lands level with Seedream 5.5, which renders larger images but only at 2K. MAI-Image-2.7 Flash drops below Seedream 5.0 Flash and sits a tenth of a cent above FLUX 3 Image Klein, which makes it one of the cheapest image models on the node that also edits from five references.
What you actually pay on Treza
On Treza, MAI-Image-2.7 is metered from a prepaid credit balance: one credit is $0.01, packs start at $5, credits do not expire, and there is no subscription. Before a run the node shows its estimate, which rounds each tier up slightly, to $0.036 an image on MAI-Image-2.7 and $0.012 on Flash, plus $0.014 and $0.005 for each reference, before the credit markup. The charge then settles on what the provider billed, and a failed generation charges nothing.
Calling MAI-Image-2.7 as an API
Calling it directly means provider credentials, the chat completions request shape with an image modality, token billing to reconcile, and retries, all before your product does anything with the image.
Through a Treza pipeline, MAI-Image-2.7 is a node on a canvas. Place the Image Generation node, pick either tier, set the aspect ratio, wire up to five references into its reference input, and publish the graph as one versioned endpoint. Your app calls the typed /invoke route with JSON in and JSON out, or points an existing OpenAI SDK at the OpenAI-compatible route. Switch the node from Flash to the precision tier later and the endpoint your product calls stays the same.
An agent can reach it too. Treza's hosted MCP server exposes pipelines as tools that Claude and other MCP clients call, so an agent can make the frame and hand back its URL as one step of a larger task.
Putting it in a pipeline instead of a single call
A single still is rarely the whole job. The useful pattern is MAI-Image-2.7 as the first node of several: compose the key frame from the product photo, pass it to a video node as the first frame, then put narration, captions or publishing behind it. A fuller AI video pipeline adds a script step ahead of the image and reruns the whole graph on a schedule or from one API call.
The AI image generator page has the image side set up, the AI video generator page puts a video model behind one generate action, and the AI thumbnail generator covers the square and wide frames. For the generation before this one, see MAI-Image-2.6 pricing, and for image graphs in more depth, how to build an AI image generation pipeline.
Frequently Asked Questions
How much does MAI-Image-2.7 cost per image?
About 3.5 cents on MAI-Image-2.7 and about 1 cent on Flash. The listed prices as of November 2026 are $0.000034 and $0.00001 per output image token, and every image in our probes was 1024 tokens, which gives about $0.035 and about $0.010 an image.
What is the difference between MAI-Image-2.7 and MAI-Image-2.7 Flash?
They take the same inputs: a prompt, up to five reference images and seven aspect ratios plus auto. MAI-Image-2.7 is the precision tier and costs about three and a half times as much. In our probes it took 17 to 29 seconds against 11 to 12 for Flash, and returned finer surface detail on the same prompt.
Is MAI-Image-2.7 cheaper than MAI-Image-2.6?
Yes. The precision tier lists at $0.000034 per output image token against $0.000038, about 10 percent less. Flash lists at $0.00001 against $0.000019 for 2.6 Flash, just under half. Both generations stay available on the node.
Do reference images cost extra?
They bill as input image tokens. Our 1280 x 853 reference read as 1,659 tokens, about 1.3 cents on MAI-Image-2.7 and about 0.4 cents on Flash at the listed rates. Up to five references go on one request.
Can MAI-Image-2.7 put readable text in an image?
Do not count on it. In our probe the precision tier garbled one word on a chalkboard while Flash wrote its words cleanly. For a title or a label, generate the image without it and add the words with the Overlay or Captions node.
Can I use MAI-Image-2.7 in chat?
Yes. Ask for MAI-Image-2.7 or MAI-Image-2.7 Flash by name in Treza chat, attach reference images if you have them, and the image comes back in the conversation. Both tiers are also in the Image Generation node's picker on the canvas.


