GPT Image 2 API Pricing: Routes, Cost Per Image, and Budgets

Current GPT Image 2 prices on HiAPI

As of August 31, 2026, HiAPI publishes five GPT Image 2 route lines: text-to-image standard, beta, and ext, plus image-to-image standard and ext. The public pricing API is the live billing source of truth; the tables below were checked against it on that date.

Try the model: Open GPT Image 2 on HiAPI to compare the available routes in the Playground and run a real test before budgeting a production batch.

The public pricing payload normalizes routed lines with names such as gpt-image-2/text-to-image@ext. In request examples, the text-to-image docs and image-to-image docs use the base slash ID plus a top-level route field. Both forms can select the route, but this guide uses the documented base-ID-plus-route shape so the different input fields remain obvious.

Standard text-to-image and image-to-image

The current standard routes have the same fixed price by resolution.

The standard text-to-image request uses prompt, aspect_ratio, and resolution. The standard image-to-image request adds input_urls, with one to six reference images. The current public pricing policy for the standard image-to-image line varies by resolution and does not publish a separate reference-image add-on.

Standard is the simplest line to forecast. A 500-image run costs $15 at 1K, $20 at 2K, or $30 at 4K. Aspect ratio does not appear as a separate billing rule, although the model docs still define which ratio and resolution combinations are valid.

The beta text-to-image route

The beta route is currently $0.02 per image. It has a different input shape from standard: pass route: "beta" at the top level and use input.size instead of input.aspect_ratio and input.resolution.

{
  "model": "gpt-image-2/text-to-image",
  "route": "beta",
  "input": {
    "prompt": "A clean editorial product photograph",
    "size": "1024x1024"
  }
}

At the current flat rate, 500 beta outputs cost $10. Treat beta as its own route contract rather than swapping only the price in a standard-route request.

Ext text-to-image output pricing

The ext text-to-image route requires quality and resolution, supports the documented multi-ratio 1K, 2K, and 4K combinations, and uses this output price matrix:

Quality is the largest single price lever in this matrix. For example, 100 text-to-image outputs at low/1K cost $0.70, while 100 at high/4K cost $76.00. Those configurations are different products for budgeting purposes, even though they share the GPT Image 2 family name.

Ext image-to-image pricing and reference fees

The ext image-to-image route uses the same output matrix as ext text-to-image, then adds $0.015 × number of reference images to each task. The route accepts one to six references in input.image_urls.

The unit-cost formula is:

ext image-to-image unit cost
= selected quality/resolution output price
+ ($0.015 × reference-image count)

For a low/1K edit, one reference image makes the unit cost $0.007 + $0.015 = $0.022. Six references make it $0.007 + $0.090 = $0.097. At 100 outputs, those two workflows cost $2.20 and $9.70 respectively.

The reference-image fee applies to ext image-to-image in the current pricing payload. Do not add it to standard image-to-image or text-to-image estimates unless the live pricing API publishes a matching add-on later.

What actually changes the bill

Route and output configuration come first. Standard is priced by resolution, beta is currently flat, and ext combines quality with resolution. A default copied from one route can therefore produce the wrong estimate for another route.

Reference-image count matters on ext image-to-image. A six-reference fusion request carries six add-on units on every output, so trimming references that do not contribute to the edit can reduce cost without changing the selected output tier.

Output count includes intentional variants and successful reruns. If a team requests four alternatives for every catalog item, the budget needs four paid outputs per item. Retry code should distinguish invalid requests, transient failures, and user-requested rerolls so it does not create duplicate successful generations.

Resolution and quality should follow the final delivery surface. Use a lower-cost configuration while validating the prompt, then run the selected prompt at the final route, quality, and resolution. This is a workflow decision, not an automatic discount.

Callbacks change orchestration, not the published unit price. A final callback reduces polling and makes terminal-state handling easier, but the current pricing policies do not make a callback-generated image cheaper than a polled image.

What does not lower the published unit price

The current HiAPI pricing payload does not expose a prompt-cache discount for these task routes. Reusing a prompt template may improve consistency and development speed, but it does not change the published per-image rule.

The payload also does not publish a lower batch unit price. Submitting related tasks together can simplify queues, concurrency, and review, but 100 outputs are still priced as 100 outputs under the current rules.

This distinction matters because generic token-API advice does not automatically apply to an async image task API. Only count a saving when it appears in the actual pricing policy or reduces a billed input such as ext reference-image count.

A production budget checklist

Before multiplying a headline price by monthly volume, answer these questions:

1. Is the job text-to-image or image-to-image?
2. Is the request using standard, beta, or ext?
3. Which resolution applies, and does the route also require quality?
4. For ext image-to-image, how many reference images will each output use?
5. How many variants and successful reruns should be included per final asset?
6. Has the public pricing API changed since this article's checked date?

For live numbers across the catalog, use HiAPI Pricing. For cost planning across image volume, open the image API cost calculator. For the model and Playground, use the GPT Image 2 model page.

FAQ

How much does GPT Image 2 cost per image on HiAPI?

On August 31, 2026, standard text-to-image and image-to-image cost $0.03 at 1K, $0.04 at 2K, and $0.06 at 4K. Beta text-to-image is $0.02 flat. Ext output prices range from $0.007 to $0.760 by quality and resolution, with an additional $0.015 per reference image on ext image-to-image.

Why does the pricing API show @ext or @beta?

That suffix identifies a priced route. The documented request shape uses the base slash model ID and a top-level route, such as model: "gpt-image-2/text-to-image" with route: "ext". HiAPI also accepts the route-qualified model form.

Is image-to-image always more expensive than text-to-image?

No. Standard text-to-image and standard image-to-image currently share the same resolution prices. Ext also shares the same output matrix, but ext image-to-image adds $0.015 for each reference image.

Does prompt caching reduce GPT Image 2 cost on HiAPI?

Not under the current public task-pricing rules. Prompt reuse can improve consistency, but the pricing payload does not publish a prompt-cache discount for these routes.

Is there a batch discount?

The current public pricing payload does not publish one. Batching can improve orchestration, but it does not lower the stated per-output price.

How often should I recheck these numbers?

Check the public pricing API before a large run and before publishing any static budget. This article records the checked date and links to the live pricing source so later changes can be detected and updated.

Next steps

• Try GPT Image 2 in the model Playground.
• Read the text-to-image API parameters and examples.
• Read the image-to-image API parameters and examples.
• Check the live HiAPI pricing catalog.
• Estimate an image-generation budget.