How Much Does GPT Image 2 Cost?
GPT Image 2 is reported to cost $5 per million text input tokens, $8 per million image input tokens, $2 per million cached image input tokens, and $30 per million image output tokens. No flat per-image price has been published. These rates come from secondary coverage rather than a pricing page we were able to confirm on OpenAI's own site, so treat them as reported, not verified.
That caveat matters. Google's competing image model, covered below, publishes a per-image price range alongside its token rates. GPT Image 2 does not, so the number you actually care about, cost per finished image, has to be derived rather than looked up.
| What you are billed for | Reported rate per million tokens |
|---|---|
| Text input | $5.00 |
| Image input | $8.00 |
| Cached image input | $2.00 |
| Image output | $30.00 |
Separate the three lines that matter. Text input at $5 per million is your prompt, a rounding error at any realistic length. Image input at $8 per million is the cost of sending a reference or source image in, which dominates editing workflows. Image output at $30 per million is the generated picture, which decides the bill in generation workflows.
The cached image input rate is the quietest line in the table and the most useful. At $2 per million against $8 uncached, resending the same reference image is 75 percent cheaper once cached, which follows directly from the two reported rates.
GPT Image 2 is positioned as the next generation image generation and editing model and the successor to GPT Image 1. Free credit programs can cover this bill outright for teams that qualify, and AI Perks tracks which providers are currently running them.
Why Token Billing Makes GPT Image 2 Hard to Budget
Your bill scales with how many output tokens each image consumes, not with how many images you generate, because a larger or higher quality image consumes more output tokens at the same reported $30 per million rate.
The honest way to budget is to work backwards from the token count. Fix the rate at the reported $30 per million and the arithmetic is simple:
| Image output tokens | Cost per image | Cost per 1,000 images |
|---|---|---|
| 500 | $0.015 | $15 |
| 1,000 | $0.030 | $30 |
| 2,000 | $0.060 | $60 |
| 4,000 | $0.120 | $120 |
| 8,000 | $0.240 | $240 |
Every figure there is arithmetic on the reported rate, not a published price. If the $30 rate is wrong or out of date, every row moves with it.
More importantly, we could not verify any published mapping from image size and quality setting to output token count. Without it, nobody can tell you what a single 1024 by 1024 image costs, including the people confidently posting per-image figures. The only reliable method is empirical: generate a small batch at your real settings, read the token counts back from the API usage object, then multiply.
Two consequences follow. Prompt length barely matters, since text input is six times cheaper per token than image output. And if your product edits images rather than generating them, your cost centre is image input, where caching is the lever.
When Was GPT Image 2 Released?
The launch date is genuinely unsettled. Most coverage dates GPT Image 2 to 24 June 2026, while at least one account places it in April 2026. We were not able to confirm either against an OpenAI announcement, so the defensible answer is mid 2026, exact date unconfirmed.
Secondary sources copy each other, so a launch date that appears in ten articles is not ten confirmations. What is consistently reported is the positioning: a next generation image generation and editing model that succeeds GPT Image 1. Claims about later variants, dated snapshot identifiers, and successor models are circulating, and none survived verification here, so this page does not repeat them.
Is GPT Image 2 Free to Use?
Yes in ChatGPT, on the free tier, with rate limits. The size of those limits is not published in any source we could verify, so plan around a handful of images per day rather than a specific documented number.
Any precise cap you see quoted, two per day or three per rolling 24 hours, is community observation rather than vendor policy, and such limits change without notice. Use the free tier to evaluate quality, not to run production volume.
On the API side, whether OpenAI operates a free allowance for image generation is not confirmed by the sources behind this page, so the safe assumption is that you pay per token from the first request. That leaves a short list of levers:
- Cached image input at a reported $2.00 per million against $8.00 uncached, a 75 percent reduction on the input side, and the biggest lever available in an editing pipeline.
- Batch processing. Google publishes a 50 percent batch discount on its competing image model. Whether OpenAI offers an equivalent on GPT Image 2 is not confirmed here, so check before building a pipeline around it.
- Free credits from provider startup programs. OpenAI runs a general startup program, and credit amounts and eligibility vary by profile rather than being published as a fixed figure. That is the category AI Perks exists to map.
GPT Image 2 vs Nano Banana 2: The Real Price Comparison
Nano Banana 2 charges $60 per million image output tokens, double GPT Image 2's reported $30, but $0.50 per million on input against OpenAI's reported $8 for images. Which one is cheaper depends entirely on whether you generate or edit.
| GPT Image 2 (reported) | Nano Banana 2, Gemini 3.1 Flash Image (official pricing page) | |
|---|---|---|
| Text input /M | $5.00 | $0.50 |
| Image input /M | $8.00 | $0.50 |
| Image output /M | $30.00 | $60.00 |
| Published per-image price | None | $0.045 at 0.5K to $0.151 at 4K |
| Batch discount | Not confirmed | 50 percent |
| Free consumer access | ChatGPT free tier, rate limited | Gemini app and Google AI Studio |
| Launch date | Mid 2026, unconfirmed | Around May to June 2026, unconfirmed |
Note the asymmetry in how much each column can be trusted. Google's rates sit on its official Gemini API pricing documentation, including a per-image range derived from resolution. OpenAI's come from secondary coverage. Same table, two confidence levels.
The input gap is the finding most comparisons miss. Google is 16 times cheaper on image input tokens, $0.50 against a reported $8.00. That is the dominant cost in any workflow that repeatedly sends reference images for editing rather than generating from a prompt, and the ratio is large enough to change an architecture decision. If your product generates from prompts instead, OpenAI's half price output wins.
How to Cut the Image Generation Bill
Check what free credits you qualify for at getaiperks.com before you put a card on file. Provider credits apply to exactly the API spend these token rates produce, so securing them is the only lever that takes the bill to zero rather than merely reducing it.
Measure before you model. Generate a batch at production settings and read the real token counts. Every per-image figure you have seen, this page included, is a guess until you do.
Cache your reference images. Editing pipelines that resend the same source image pay the reported $8 per million instead of $2 until they do, four times the rate on what is usually their dominant cost.
Match the model to the workload. Editing heavy products should price Google's $0.50 input against OpenAI's reported $8, and generation heavy products should compare OpenAI's reported $30 output against Google's $60.
| Program | What it is | Where to check |
|---|---|---|
| OpenAI for Startups | OpenAI's general startup program, credit amounts vary by profile | AI Perks |
| Google for Startups Cloud Program | Google's general startup cloud program, credit amounts vary by profile | AI Perks |
| Other provider programs | Most major AI and cloud providers run comparable credit programs | AI Perks |
Frequently Asked Questions
How much does GPT Image 2 cost per image?
No per-image price has been published. Billing is reported as token based at $30 per million image output tokens, so an image consuming 1,000 output tokens works out to $0.03 and one consuming 4,000 works out to $0.12. Those are calculations on a reported rate, not quoted prices. Generate a test batch at your real settings and read the token counts from the API usage object.
Is GPT Image 2 free?
It is free to use in ChatGPT on the free tier, subject to rate limits that are not published in any source we could verify. Precise figures circulating for that cap are community reports rather than vendor policy. A free API allowance for image generation is not confirmed, so the way to zero an API bill is provider credits, tracked at getaiperks.com.
When was GPT Image 2 released?
Mid 2026, with the exact date unconfirmed. Most coverage says 24 June 2026 and at least one account says April 2026, and neither could be verified against an OpenAI announcement here. If the precise date matters for your work, confirm it with the vendor.
How do GPT Image 2 and Nano Banana 2 compare on price?
It depends on the workload. GPT Image 2 is reportedly half the price on image output, $30 against $60 per million tokens. Nano Banana 2 is 16 times cheaper on image input, $0.50 against a reported $8.00. Generation favours OpenAI, editing favours Google.
Are the prices on this page official?
Google's figures come from its official Gemini API pricing documentation. OpenAI's figures come from secondary coverage rather than a confirmed OpenAI page, which is why they are described as reported throughout. Verify against the vendor before committing them to a financial model.
Can I get free credits that cover image generation?
Provider and cloud startup programs do issue API credits that cover image model spend, and the amounts vary by profile rather than being published as a fixed figure. OpenAI and Google both run general startup programs of this kind. AI Perks tracks what is currently available.
Generate the images. Let someone else pay for the tokens.