GPT Image 2.6 Pricing: Sunburst, Flare and Economy (2026)

GPT Image 2.6 pricing has two different units hiding behind familiar model names. OpenAI and KIE list Sunburst and Flare by tokens. KIE also offers an Economy route at a fixed price per 1K image. Before deciding which is cheaper for your profile photos, identify which unit the service is selling and what work that price includes.

The Ultimate Profile team checked the primary pricing and model documentation on 5 November 2026 and made one small stock-photo editing call through each KIE route. The guide below distinguishes published rates, measured sample debits and hypothetical calculations. None of those numbers is a promise about the cost or quality of an unseen future image. For the release itself, start with the Images 2.6 profile-photo and upgrade guide.

The pricing framework: tokens, images and customer credits

Three units matter. Tokens measure the inputs and output a provider bills. A per-image tariff assigns a price to an eligible generation under defined settings. Customer credits belong to the website where you buy or rerender a picture. The word credit can appear in both provider and customer interfaces without representing the same amount of money or the same entitlement.

OpenAI's official pricing lists the API variants gpt-image-2.6-sunburst and gpt-image-2.6-flare. Both have the same published token rates. KIE's Sunburst and Flare entries also share rates, while its separate Economy route is listed at 5 KIE credits per 1K image. A KIE credit represents approximately $0.005 on that pricing page; it is not an UP Credit in our shop (OpenAI pricing, KIE pricing, checked 5 November 2026).

An advertised price can therefore answer one question while leaving another unresolved. A token rate tells you how to price a known token count. It does not reveal the token count of the portrait you have not generated. A fixed image price can be easier to predict, but only within the route's supported settings. A customer credit price tells you the debit on that website, which may fund more than raw model access.

Keep a fourth concept separate: the cost of an accepted picture. If you pay for several generations before selecting one, your total spending covers those attempts. A low unit price does not automatically make an editing process economical when the outputs do not meet your acceptance criteria. That is a practical accounting distinction, not a measured success-rate claim about any tier.

Unit	What the number describes	What else you need
Provider tokens	Billable input and output usage	Actual usage and applicable rates
Provider image tariff	One eligible generation	Supported size, mode and conditions
Website credits	Your debit in that product	The displayed price and entitlements

When comparing two offers, rewrite each in the same terms: what do you pay, what do you receive, and what happens if you need another attempt? That small exercise prevents a token price, a subscription allowance and a fixed rerender price from being treated as three quotes for an identical service.

Sunburst vs Flare: identical rates do not fix the image bill

OpenAI's current pricing page lists both 2.6 variants at $5 per million text-input tokens, $8 per million image-input tokens and $30 per million image-output tokens. Cached text input is $1.25 per million tokens, and cached image input is $2 per million. These are the official published units, read on 5 November 2026. They should not be restated as prices for one generated image.

The rate equality matters because the names can suggest an unsupported commercial hierarchy. Sunburst is described for detailed work and stronger consistency across edits. Flare is described for faster iteration. That positioning does not mean Sunburst necessarily has a higher rate per token. It also does not guarantee the same token usage, final debit or time to reach an accepted portrait for every task (OpenAI release, checked 5 November 2026).

For a profile-photo edit, the inputs can include a text instruction and existing pictures. The output also consumes billable usage. A comparison that includes only the image-output rate leaves out the input side of the work. This matters particularly when you move from generating a fictional portrait from text to editing a real reference photograph. Those are different input conditions even if the requested output looks similar.

Caching introduces another distinction. The pricing table contains cached-input rates, but their presence does not prove that a specific request will be billed as cached. Do not write a budget that silently assumes all references receive a cache discount. Look for actual usage information or a clear contract for the service you are using. Reusing a file name or sending the same URL is not, by itself, a billing receipt.

Official rate per million tokens	Sunburst	Flare
Text input	$5	$5
Cached text input	$1.25	$1.25
Image input	$8	$8
Cached image input	$2	$2
Image output	$30	$30

Choose between the variants based on the documented workflow and your actual output requirements. If cost is the deciding factor, compare the billed usage or the website's displayed fixed charge for that workflow. The shared rate table answers which rates apply; it does not declare an automatic winner for every portrait you might ask either variant to create.

KIE token pricing and a calculation you can audit

KIE lists Sunburst and Flare at $4.50 per million text-input tokens, $7.20 per million image-input tokens and $27 per million image-output tokens. Its cached-input prices are $1.125 for text and $1.80 for images. In KIE credits, those rates are respectively 900, 1440, 5400, 225 and 360 credits per million tokens. Both tiers share the same published rates (KIE pricing, checked 5 November 2026).

The arithmetic is straightforward when the usage is known. Multiply each applicable token count by its rate, divide by one million and add the components. Keep cached and uncached input separate, because assigning the wrong category changes the result. The difficult part for a buyer is often obtaining the actual usage rather than performing the multiplication.

Consider a deliberately hypothetical example with 100 uncached text-input tokens, 1000 uncached image-input tokens and 1000 image-output tokens. At KIE's published rates, those components would cost $0.00045, $0.00720 and $0.02700. The total would be $0.03465, or 6.93 KIE credits. These token counts were chosen only to explain the formula. They are not measured portrait usage, a typical image bill or a forecast for your next request.

If the image input were actually billed as cached, that component would use the cached rate instead. The word actually matters: a cheaper hypothetical calculation does not establish that the discount occurred. A useful receipt distinguishes the usage categories or at least provides the final bill. Keep the distinction visible when a tool displays an estimate before generation and a different amount afterward.

KIE rate per million tokens	USD	KIE credits
Text input	$4.50	900
Cached text input	$1.125	225
Image input	$7.20	1440
Cached image input	$1.80	360
Image output	$27	5400

For an ordinary website user, you may never see token counts. In that case, focus on the price the website commits to charge and the settings it covers. A product can absorb provider variation behind a fixed customer price, but that is the product's pricing decision. You should not have to reconstruct a private provider invoice to know what pressing the rerender button will debit from your wallet.

Economy is a fixed 1K route with a narrower contract

KIE's GPT Image 2.6 Economy entry lists 5 credits, or $0.025, per 1K image for text-to-image and image-to-image. Its model page describes an economy channel with a flat per-image price. The primary OpenAI pricing page we checked names Sunburst and Flare; it does not provide a separately named Economy API row. Describe Economy as the KIE access route you are buying rather than inventing a third official OpenAI model identifier.

The narrower contract is part of the comparison. KIE's Economy editing documentation supports PNG output and 1K generation. Its aspect-ratio list includes auto, 1:1, 16:9, 9:16, 3:2, 2:3, 21:9 and 4:3. Its editing request can take reference-image URLs. Sunburst and Flare have broader documented size and format options, so a task requiring a particular unsupported option cannot be solved by comparing Economy's price alone (KIE Economy documentation, checked 5 November 2026).

For profile-photo readers, the practical question is whether a 1K PNG in a supported frame meets the intended use. Larger output can be relevant when you need a particular delivery size, but a larger file does not independently establish stronger identity preservation. Decide what your destination needs before paying for size as a substitute for reviewing the face.

The fixed tariff also changes how you plan revisions. If each eligible attempt has the same stated price, you can calculate a maximum generation budget from the number of attempts you choose to allow. That makes spending easier to predict. It does not mean you should generate more variations automatically, or that any set number of attempts guarantees an acceptable result.

KIE route	Billing unit	Documented size choices	Buyer implication
Sunburst 2.6	Tokens	1K, 2K, 4K	Usage affects the provider bill
Flare 2.6	Tokens	1K, 2K, 4K	Shared rates do not fix usage
Economy 2.6	Per eligible image	1K	Predictable tariff within its contract

The cheapest suitable route is the one that covers the task without charging for requirements you do not have. Suitability includes available controls, an acceptable output and a price you understand. It does not follow from a tier name alone. If the narrow Economy contract fits your still portrait, it is a reasonable route to consider; if a required format is missing, choose a route that actually provides it.

What our three real API calls establish

The Ultimate Profile team made three KIE image-to-image calls on 5 November 2026: one Economy, one Sunburst and one Flare. Each used the same public stock portrait and a narrow instruction to retain the adult's face while changing the background and light. The requested frame was 2:3 at the offered 1K size. All three returned 1024 by 1536 PNG files, which we downloaded and inspected.

The completed Economy task reported 5 KIE credits. Sunburst and Flare each reported 7.65 credits, or $0.03825. The total debit was 20.3 KIE credits, equivalent to $0.1015. These are retained integration observations for those three requests. They are not an average cost, a worst-case price, a quality ranking or evidence about every face and editing request.

The returned task details included credit consumption but no billed token breakdown. The result objects contained image URLs rather than usage categories. We therefore cannot report measured input, cached-input or output-token counts for these calls. Dividing the total debit by the image-output rate would invent a token count by ignoring the input side of the bill.

The distinction affects what can be responsibly priced. A fixed Economy tariff can be checked against the observed Economy debit. A token-billed sample can show that a request completed at a particular cost, but it cannot establish a ceiling for a future request with more references or a different output. One completed portrait does not resolve that uncertainty.

One-request 1K observation	KIE credits	USD	Interpretation
Economy	5	$0.025	Matches the fixed published tariff
Sunburst	7.65	$0.03825	Sample debit under token billing
Flare	7.65	$0.03825	Sample debit under token billing

These small calls are useful because they verify the exact request route and returned output rather than assuming that 2.5 naming still works. They do not justify a headline that a particular tier always costs a certain number of cents, preserves every face or produces better profile outcomes. Keep the sample's modest scope beside its numbers whenever you use it to inform a purchasing decision.

Compare the cost of an accepted portrait, not only an attempt

The price that matters to your budget includes the work you actually buy. If you purchase several generations and select one, account for all the paid attempts. If a service includes renders in another purchase, read what is included before adding rerender prices on top. These are different commercial arrangements, and comparing them requires a clearly defined deliverable.

Define that deliverable before browsing galleries. You may need one current professional headshot with a plain background, or one casual portrait for a social profile. A package of scene concepts, reference handling and included generations can be useful when you want that workflow. Raw API access may suit somebody who already has a way to prepare inputs and review outputs. The nominal provider charge does not tell you which arrangement saves you effort.

Use a stopping rule. Decide what makes the image acceptable and how much you are willing to spend trying to obtain it. If the face changes, do not count the attempt as a success because the scenery looks polished. Return to a stronger source or a narrower editing request before purchasing more variations of the same unsuitable task. The 2.6 editing workflow explains that decision process without promising a success rate.

Retain enough information to compare your own sessions. Note the service, tier, requested size, visible charge, total attempts and the image you accepted. You do not need to publish personal photographs or account details to keep useful private records. A simple log can show whether an apparently inexpensive route repeatedly requires work you did not budget for.

Buying question	Useful record	Decision it supports
What did the session cost?	All paid attempts	Total spending rather than favorite-image price
What was included?	Purchase terms and displayed price	Avoiding double-counted charges
Why was an output rejected?	A specific visible defect	Improving the next request
When should you stop?	Acceptance rule and spending limit	Avoiding endless variations

Do not turn a personal log into a universal model ranking. Your face, source image, task and tolerance for revision are part of the result. Another reader may need a different output. The useful conclusion is narrower and actionable: you know what a particular workflow cost you to obtain a picture you were willing to use.

Before you pay: match the offer to the route

Read the generation screen as a contract for the next action. Confirm the selected model or channel, the size, the output type and the debit shown beside the button. If a website uses its own credits, compare the wallet debit with the credits you actually receive from your purchase. Promotional credits and provider credits are separate units, so a matching number in two interfaces is not evidence of matching cash value.

Check whether an action creates a new paid attempt or merely changes a preference. A model picker can save your choice for later without generating anything. A rerender can create a new output and debit the amount displayed for that model and size. Keeping those actions distinct helps you avoid buying a variation when you meant only to inspect the available settings.

Read the failure and retry terms provided by the service. Do not assume that every unsuccessful visual result receives a refund, or that a technically failed task and a photograph you dislike are treated the same way. Those questions concern the website's commercial rules. A model's published token table cannot answer them.

Check whether an older article describes the same pricing route. Our 2.5 Sunburst and Flare guide records a different model generation and dated conditions. Even when model names resemble one another, retain the version and access-route labels. The 2.6 token tariffs should not be replaced with a 2.5 reseller's historical per-image table.

Finally, keep the displayed price associated with the image you requested. If the service quotes a fixed customer debit, the completed action should follow that quote. If it offers a usage estimate instead, understand that distinction before starting. A predictable buying decision begins with knowing whether you are accepting a fixed price, a credit allowance or a bill that depends on usage.

Match generation and editing to the picture you need

A text-to-image tariff and an image-to-image tariff can look similar while serving different needs. If you want a picture of yourself, an existing reference is part of the task. Generating a fictional person from a written description cannot demonstrate that the service preserved your identity, even if the fictional portrait is inexpensive and visually convincing. Choose the mode that actually receives the information the job requires.

An editing route also deserves a careful input check. Confirm that the intended source is attached, that the service shows the correct reference order when several images are present, and that the requested change is explicit. Paying for an output made from the wrong reference is still spending money, regardless of how low the quoted generation price was. A brief pre-generation check can prevent an avoidable attempt without claiming that it improves the model itself.

Keep this distinction when reading example galleries. A text-generated portrait can illustrate rendering style. An edited reference can illustrate a particular transformation. Neither example reveals the full cost of obtaining your own acceptable photo. Ask which mode produced the example before using its displayed price to estimate your task.

Where The Ultimate Profile fits

We sell an AI photo product, so treat this section as interested. Our shop adds GPT Image 2.6 Economy at 1K for 7 UP Credits per paid rerender. That is our customer price, not 7 KIE credits and not a claim that the provider charges $0.07. The documented provider tariff is 5 KIE credits, or $0.025. Existing set purchases retain their included-render rules.

Sunburst and Flare 2.6 are not selectable in this update. Their token billing needs a verified way to bound costs before it can fit the shop's fixed displayed prices. We do not convert the 7.65-credit sample debits into pretend fixed tariffs. Existing models and prices remain available, and Nano Banana 2.1 remains the default.

This choice addresses predictable billing, not an independently established quality ranking. The three stock-photo calls do not establish improved likeness from one selfie and do not support claims about more matches, replies or interviews. Judge the resulting portrait against your own source and intended use.

Choose a price you can explain before generating

For GPT Image 2.6, separate the token rate, the generation tariff and the customer debit. Sunburst and Flare share rates, while actual usage still matters. KIE Economy offers a narrower fixed 1K route. A website can package those costs differently, so read the offer where you will actually press the button.

Choose the route that supplies the controls and file you need, keep a record of the full session cost, and stop when an accepted portrait meets your criteria. A clear price does not guarantee a suitable photograph. It does let you decide how much editing you are willing to buy without confusing one provider's accounting unit with another product's promise.