MAI-Image-2.6 API and Pricing, Explained Through a Real Pipeline

MAI-Image-2.6 is Microsoft AI's image generation and editing family, and it comes in two tiers that take the same inputs. MAI-Image-2.6 is the precision tier, built for design-ready visuals. MAI-Image-2.6 Flash is the lower-latency, lower-cost tier for high-volume work. Both generate from a prompt alone, and both edit or compose from a prompt plus up to five reference images, in seven aspect ratios plus auto. We put both on the Image Generation node, in chat, and on a model page on October 14, 2026, after probing them through the same path the node uses.

The pricing question has a wrinkle, because the rate card is written per output image token rather than per image. This post covers that rate card, how it turns into a flat per-image price, what our test renders billed and how long they took, how the Flash reference edit behaved, and how to call either tier as a published API endpoint instead of writing a Microsoft integration.

The spec sheet

Every row is the same on both sides. The seven ratios are 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3, which covers a square thumbnail, a vertical story frame and a wide header from one node, and leaving the ratio on auto lets the model pick one from the prompt. There is no resolution setting on either tier: in our tests at 1:1 both came back as 1024 pixel squares.

Because the inputs match, a pipeline built on one tier switches to the other by changing the model on the node and nothing else.

What MAI-Image-2.6 costs per image

These are the listed prices as of October 2026. Both tiers are priced per output image token:

The second column is the one you plan with. Every image in our probes was exactly 1024 image tokens, so the per-token rate becomes a flat price: about 3.9 cents on MAI-Image-2.6 and about 2 cents on Flash, with Flash at precisely half the rate. Reference images bill as input tokens on top, and in our edit one reference added less than half a cent.

For scale: a hundred Flash images list at about $1.95. The same hundred on MAI-Image-2.6 list at about $3.90. On the Treza node the credit estimate shown before a run rounds each of those up slightly, to $0.04 on MAI-Image-2.6 and $0.021 on Flash, and the charge settles on what the provider billed.

The family also has an older generation in the catalog. MAI-Image-2.5 lists at $0.000047 per output image token and MAI-Image-2.5 Pro at $0.000108, so MAI-Image-2.6 is cheaper than the model it follows, and Flash is cheaper than anything the family has offered before.

What we measured on October 14

We probed both tiers on October 14, 2026, through the same chat completions path the Image Generation node uses. The billed cost matched the rate card each time:

The text prompt, the same on both tiers, asked for a ceramic cup of coffee on a sunlit wooden cafe table with steam rising and shallow depth of field. MAI-Image-2.6 returned a speckled stoneware cup on worn wood, a small vase of white flowers, a blurred stone street behind, and a notebook and pen beside the cup with three handwritten words on the cover that the prompt never asked for. Flash returned a cream cup on a saucer with two stacked books and a potted plant beside it. Both are clean, plausible photographs. The precision tier spent its extra fourteen seconds on finer surface detail, the glaze speckle and the wood grain.

The edit took a product still of a folded navy linen shirt and asked Flash for the same shirt on a white marble counter in a bright studio. The model kept the shirt, its weave, collar, button and fold, set it on veined marble under hard sunlight, and came in close on the fabric. The product survives and the scene changes, which is the behaviour you want from a reference edit.

The edit's $0.0238 breaks down cleanly: 1024 output tokens at the Flash rate, $0.0195, plus about 1,800 prompt tokens for the reference, $0.0044. So a reference on Flash costs under half a cent. Two renders per tier on one day is not a benchmark, but the gap was consistent: Flash in ten to fifteen seconds, MAI-Image-2.6 in twenty-four.

Flash to draft, MAI-Image-2.6 to finish

With two tiers on one spec sheet at a two-to-one price ratio, the pattern writes itself:

1. Draft on Flash. Explore framing and prompt wording at about 2 cents a frame and ten seconds a render. Twenty-five variations list at about $0.49.
2. Finish on MAI-Image-2.6 when a frame earns it. Rerun the chosen prompt on the precision tier for about 4 cents, once rather than twenty-five times. The prompt, the ratio and the references carry across unchanged.
3. Keep references on both. Unlike models that charge a cent per input image, a reference here is a fraction of a cent, so there is no reason to leave the product out of the drafts.

On the Treza node that is one change in the model menu. In chat, ask for Flash by name while you explore, then ask for MAI-Image-2.6 on the keeper.

Reference images and the edit path

With a prompt alone either tier generates. With a prompt plus up to five references it edits or composes, on the same node. Wire the images into the Image Generation node's reference input, or drag them onto it, and write the change as the prompt: the same shirt on a marble counter, or this logo on a tote bag.

Five references is a middle ceiling in the catalog. Nano Banana 2.1 takes up to 14 and FLUX 3 Image Klein up to 10, while Grok Imagine Image 2.0 takes three. For a product, a setting, a style and a logo it is enough. For a brief that has to carry a dozen assets into one frame, pick one of the larger ceilings instead.

Where it sits in the price band

Listed and measured prices as of October 2026 for the image models around it, at 1K where the model has a resolution setting:

MAI-Image-2.6 lands in the same band as Nano Banana 2.1 and the Grok low tier, and Flash lands a hair above Seedream 5.0 Flash and Qwen Image 2.1 Turbo. Neither is the cheapest image model on the node. What the pair offers is a precision tier and a speed tier on one spec sheet, five-reference editing on both, and a reference charge small enough to ignore. For design-ready stills and product edits where surface detail is the point, MAI-Image-2.6 is the tier to finish on. For a storyboard of fifty frames where the count matters more than the look, Seedream 5.0 Flash or FLUX 3 Image Klein will do it for less.

What you actually pay on Treza

MAI-Image-2.6 through a Treza pipeline is metered from a prepaid credit balance. One credit equals $0.01, packs start at $5, credits never expire, and there is no subscription. A failed generation charges nothing, and that includes a prompt the provider refuses. Treza quotes the price before a run, rounding the per-token rate up to the nearest tenth of a cent, and settles the charge on what the provider billed.

Calling MAI-Image-2.6 as an API

Direct access means Microsoft credentials, the request shape of the chat completions route with an image modality, token-level billing to reconcile, and retry logic before you have built anything.

Through a Treza pipeline, MAI-Image-2.6 is a node on a canvas. Place the Image Generation node, pick either tier, set the aspect ratio, wire references into its reference input, and publish the graph as a single 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. Swap the node from Flash to the precision tier later and the endpoint your product calls stays the same.

The same pipeline is reachable from an agent. Treza's hosted MCP server exposes pipelines as tools that Claude and other MCP clients call directly, so an agent can generate the frame and hand back the URL as one step of a larger task.

Putting it in a pipeline instead of a single call

A single image is rarely the deliverable. The more useful pattern is MAI-Image-2.6 as the first node of several: compose the key frame on the precision tier, pass it to a video node as the first frame, then put narration, captions, or publishing behind the video. A fuller AI video pipeline puts a script step ahead of the image and reruns the whole graph on a schedule or from an API call.

The AI image generator page has the image side set up, the AI video generator page puts a video model behind a single generate action, the AI thumbnail generator page covers the square and wide frames, and how to build an AI image generation pipeline walks through image graphs in more depth. For its closest neighbour on price, see Grok Imagine Image 2.0 pricing.

Frequently Asked Questions

How much does MAI-Image-2.6 cost per image?

The listed prices as of October 2026 are $0.000038 per output image token on MAI-Image-2.6 and $0.000019 on Flash. Every image in our probes was 1024 tokens, so that is about $0.039 an image on MAI-Image-2.6 and about $0.0195 on Flash, and our text-to-image probes billed exactly those amounts.

What is the difference between MAI-Image-2.6 and MAI-Image-2.6 Flash?

They take the same inputs: a prompt, up to five reference images, and seven aspect ratios plus auto. MAI-Image-2.6 is the precision tier and Flash is the faster, cheaper one at half the per-token rate. In our tests MAI-Image-2.6 took about 24 seconds and Flash about 10, and the precision tier returned finer surface detail on the same prompt.

Do reference images cost extra with MAI-Image-2.6?

They bill as input tokens, and the amount is small. Our Flash edit with one reference billed $0.0238 against $0.0195 for a plain prompt, so the reference added about $0.0044. Up to five references go on one request, wired into the Image Generation node's reference input.

Can MAI-Image-2.6 edit an image I already have?

Yes. Both tiers accept image input as well as text. Attach the image as a reference and write the change as the prompt. In our probe, Flash kept a folded navy shirt intact and moved it from its original still onto a white marble counter in a bright studio.

What aspect ratios does MAI-Image-2.6 support?

Seven, plus auto: 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3. There is no resolution setting on either tier. In our tests at 1:1 both returned 1024 pixel squares.

Can I use MAI-Image-2.6 in chat?

Yes. Ask for MAI-Image-2.6 or MAI-Image-2.6 Flash by name in Treza chat, attach reference images if you have them, and the image comes back in the conversation. Both tiers also sit in the Image Generation node's model picker on the canvas.