From Aryan Vatsa | Product & Market Analysis

AI Image API Pricing per 1,000 Images: GPT Image, Gemini and FLUX on One Basis

On this page

On official rate cards, 1,000 square images at roughly 1 megapixel cost anywhere from $5 to $211, depending on which vendor and which quality setting you pick. AI image API pricing looks hard to compare because OpenAI, Google and Black Forest Labs each bill on a different unit. Convert all three to the same 1,000 images at the same size and the spread becomes a 42x decision, not a rounding error.

Key takeaways

  • The same 1,000 square images cost $5 to $211 across current official rate cards. The cheapest is gpt-image-1-mini at low quality and the most expensive is gpt-image-2 at high quality, both from OpenAI's own published tables.
  • Imagen as a standalone API is gone. Google shut down the Imagen 4 model IDs on the Gemini API on 17 August 2026 and points users to Gemini Nano Banana 2.1, which costs $33.60 per 1,000 images at the 1K tier.
  • FLUX bills per megapixel and rounds up. Black Forest Labs defines a megapixel as 1,048,576 pixels, so a 1080 by 1080 social square bills as 2 megapixels and costs 50% more than a 1024 by 1024 square on FLUX.2 [pro].
  • Indemnity is not a price line, but it is the largest cost difference. OpenAI's API terms include an output IP indemnity. Black Forest Labs' developer terms run the indemnity the other way and cap its liability at the greater of $100 or six months of fees.
$5 to $211Cost of 1,000 square images at 1024 by 1024 across OpenAI's own published per-image tables. Source: OpenAI image generation guide, November 2026.
17 Aug 2026Date Google shut down all three Imagen 4 model IDs on the Gemini API. Source: Gemini API deprecations, 2026.
+50%Extra cost of a 1080 by 1080 image over a 1024 by 1024 image on FLUX.2 [pro], from megapixel rounding. Source: Black Forest Labs pricing, 2026.

The short answer

For 1,000 marketing images at about 1 megapixel, expect roughly $14 to $70 on FLUX.2, $34 to $134 on Google's Gemini image models, and $6 to $211 on gpt-image-2 depending on quality. Batch processing halves Google's price. Only OpenAI's API terms, among the three, clearly include an output IP indemnity.

This post is written for a founder or marketing director deciding which image API to standardise on for a content pipeline. The figures here are unit economics, and the decision they feed is a vendor commitment, so that is the reader they are built for.

Why AI image API pricing does not line up across vendors

Every vendor in this comparison publishes a rate card, and no two of those rate cards use the same unit. That is the entire reason a simple "which is cheaper" question takes an afternoon to answer.

The billing unit also decides which of your own choices move the bill. Under one vendor the quality setting matters most, under another it is pixel count, and under a third it is a resolution tier that ignores the exact dimensions you asked for.

Per image, per quality setting. OpenAI publishes a table of per-image prices for each GPT Image model by quality (low, medium, high) and size. Underneath, it bills image output as tokens, currently $30 per million image output tokens on gpt-image-2. The per-image table is a translation of that token count into dollars, which is helpful until the vendor stops publishing the translation.

Per token, per resolution tier. Google's Gemini image models also bill output as tokens, at between $30 and $120 per million depending on the model. The rate card then converts that into a price per image at 1K, 2K and 4K tiers. You pick a tier, not a pixel dimension, and the tier sets the token count.

Per megapixel, rounded up. Black Forest Labs prices its FLUX.2 family per megapixel, with a higher rate for the first megapixel and a lower rate for each one after it. Resolution is rounded up to the next whole megapixel, which is where the surprises live. Its older FLUX 1.1 models and newer FLUX 3 family are priced per image instead.

The matched assumptions behind the cost-per-1,000 table

A comparison is only as fair as the assumptions you hold constant. Here are the ones this post uses, stated so you can change any of them and redo the arithmetic yourself.

First, every figure is list price on the vendor's own public rate card, retrieved on 10 November 2026. No reseller prices, no volume discounts, no enterprise contracts. Your negotiated price may be lower, and that is a good reason to ask for it.

Second, the table uses three sizes that map to real marketing work. A square of about 1 megapixel covers social posts and product tiles. A 3:2 landscape of about 1.5 megapixels covers blog headers and ad creative. A 4 megapixel image covers print-quality hero shots and anything you plan to crop.

Third, the cost counts only output. Text prompts add a small input charge on OpenAI and Google, and reference images used for editing add a larger one. For a pure text-to-image pipeline with short prompts, input is a rounding error, and including it would obscure the comparison rather than sharpen it.

Fourth, Google's rate card prices by tier rather than pixel dimensions, so the landscape column uses its 1K price and the 4 megapixel column uses its 2K price. That is the nearest honest mapping, and it is flagged again in the limitation section below.

Image generation cost per 1,000 images at marketing sizes

This is the table the rest of the post explains. Each cell is the published per-image price multiplied by 1,000, in US dollars, before tax.

Cost of 1,000 images, official list prices, 10 November 2026
Model and settingSquare, about 1 MPLandscape, about 1.5 MPAbout 4 MP
gpt-image-2, low$6$5Not published
gpt-image-2, medium$53$41Not published
gpt-image-2, high$211$165Not published
gpt-image-1-mini, medium (ends 1 Dec 2026)$11$15Not offered
Gemini Nano Banana 2.1, standard$33.60$33.60 (1K tier)$50.40 (2K tier)
Gemini Nano Banana 2.1, batch$16.80$16.80 (1K tier)$25.20 (2K tier)
Gemini 3.1 Flash Image$67$67 (1K tier)$101 (2K tier)
Gemini 3 Pro Image$134$134$134
FLUX.2 [klein] 4B$14$15$17
FLUX.2 [pro]$30$45$75
FLUX.2 [max]$70$100$160
FLUX 3 Image$48 (1K)$70 (1.5K)$100 (2K)

OpenAI square is 1024x1024 and landscape is 1536x1024. FLUX.2 cells are computed from the per-megapixel rates and the round-up rule, with 1536x1024 billing as 2 MP and 2048x2048 as 4 MP. Gemini 3 Pro Image charges the same for 1K and 2K. The "where Google is cheaper" row is the batch row, and it only applies to jobs that can wait.

Cost of 1,000 square images at about 1 megapixel Official list prices, US dollars, 10 November 2026. Output only, standard tier. gpt-image-2, high$211 Gemini 3 Pro Image$134 FLUX.2 [max]$70 Gemini 3.1 Flash Image$67 gpt-image-2, medium$53 FLUX 3 Image, 1K$48 Nano Banana 2.1$33.60 FLUX.2 [pro]$30 FLUX.2 [klein] 4B$14 gpt-image-2, low$6 Sources: OpenAI, Google AI Studio and Black Forest Labs rate cards. Chart: Zan Digital.
The top and bottom bars are the same OpenAI model. Your quality setting moves the bill more than your choice of vendor does.

GPT Image pricing: tokens behind a per-image table

OpenAI's published table for gpt-image-2 prices a 1024 by 1024 image at $0.006 on low, $0.053 on medium and $0.211 on high. That is a 35x gap inside one model, and it is the most important number in GPT Image pricing. Your team's quality default matters more than the model name.

The table also contains an oddity worth knowing. At every quality level, OpenAI's gpt-image-2 table prices 1536 by 1024 landscape below 1024 by 1024 square, for example $0.165 against $0.211 on high. Every other vendor here charges more for more pixels. We report it as published, and you should test it on your own account before you plan around it.

The implied token counts, and why they matter now

Divide the published per-image price by the $30 per million token rate and you get the token count OpenAI is charging for. Our arithmetic gives about 200 output tokens for a low-quality square, about 1,767 for medium and about 7,033 for high. Those figures are derived, not published, and they are the only way to estimate cost on the newest models.

That matters because OpenAI marked gpt-image-2 deprecated on 10 November 2026, with removal from the API on 10 February 2027. The replacements, gpt-image-2.6 Sunburst and Flare, carry the same $30 per million output token rate. Their model pages publish no per-image table, and OpenAI's guide says equal token rates do not mean equal cost per image.

So here is the honest planning rule. On any GPT Image model at $30 per million, every 1,000 output tokens per image adds $30 to the cost of 1,000 images. If Sunburst spends as many tokens as gpt-image-2 medium, 1,000 square images cost about $53. Measure ten real generations before you assume it does.

Deprecation churn is a cost too

gpt-image-1 was shut down on 23 October 2026. gpt-image-1.5 and gpt-image-1-mini are scheduled for removal on 1 December 2026. That is three model retirements inside about four months on one vendor's image line.

The cheapest OpenAI row in our table, gpt-image-1-mini at $5 per 1,000 low-quality squares, has three weeks left to live. I would not build a cost model on it. Track this kind of churn the way you track price changes, because a forced migration re-tests every prompt you rely on. We keep a running record of these dates in the API price and model sunset calendar.

Imagen pricing after Imagen: what Google bills now

If you searched for Imagen pricing, the first fact is that the product you searched for has been retired on Google's developer API. All three Imagen 4 model IDs on the Gemini API were shut down on 17 August 2026. Google's Vertex AI model page lists the same three endpoints as discontinued and told customers to migrate before 30 June 2026.

The Gemini API pricing page no longer shows Imagen 4 rates at all. Any article still quoting a per-image Imagen price is describing something you cannot buy through that API today. The named replacement on the Gemini API is Gemini Nano Banana 2.1, released on 6 October 2026.

Nano Banana 2.1 is the new Imagen price point

Nano Banana 2.1 bills image output at $30 per million tokens, which Google converts to $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. That makes 1,000 standard 1K images $33.60. On the Batch API, Google halves every one of those figures.

The step up is Gemini 3.1 Flash Image at $0.067 per 1K image, and the premium option is Gemini 3 Pro Image at $0.134 for both 1K and 2K. The flat 1K and 2K price on the Pro model is useful: if you need 2K output, the 2K tier is the only one worth paying for on that model.

Batch is the cheapest lever in this comparison

The single largest discount on any rate card here is Google's 50% batch reduction. 1,000 Nano Banana 2.1 images at 1K drop to $16.80 if your pipeline can wait for results. Most marketing image work can wait, because nobody needs a quarterly campaign's product tiles in under a second.

OpenAI also halves its token rates for batch jobs on gpt-image-2, but publishes no per-image batch table. Black Forest Labs publishes no batch discount on its pricing page. If your workload is scheduled rather than interactive, test a batch run before you compare anything else, and see our breakdown of batch and caching discounts across model APIs.

FLUX API cost and the megapixel rounding trap

FLUX.2 [pro] costs $0.03 for the first megapixel and $0.015 for each additional one, per the Black Forest Labs rate card. A 1024 by 1024 image is exactly 1 megapixel under BFL's definition, so 1,000 of them cost $30. That makes FLUX.2 [pro] the cheapest production-grade option in the table at the square size.

The trap is the definition and the rounding. BFL states that one megapixel equals 1,048,576 pixels and that resolution is rounded up to the next whole megapixel. Any image even slightly larger than 1024 by 1024 bills as 2 megapixels.

FLUX.2 [pro]: cost of 1,000 images at common marketing sizes Billed megapixels shown in each column. 1 MP = 1,048,576 pixels, rounded up. $301 MP1200x628link image $301 MP1024x1024square $452 MP1080x1080social square $452 MP1920x1080video frame $754 MP2048x2048hero image
The red column is the costly mistake. A 1080 square holds 11% more pixels than a 1024 square and costs 50% more.

Why the 1080 square costs 50% more

A 1080 by 1080 image holds 1,166,400 pixels, which is about 11% more than BFL's 1,048,576 pixel megapixel. It therefore rounds up to 2 megapixels and costs $0.045 instead of $0.03. That is a 50% price increase for a size difference nobody would see on a phone screen.

The fix costs nothing. Generate at 1024 by 1024 and upscale or pad to 1080 in your own pipeline. On 100,000 social images a year, that one setting is worth $1,500 on FLUX.2 [pro] and $3,000 on FLUX.2 [max]. A 1920 by 1080 frame, by contrast, holds 2,073,600 pixels and fits inside 2 megapixels, so it bills at the same $0.045.

The cheap tier and the per-image models

FLUX.2 [klein] 4B costs $0.014 for the first megapixel and $0.001 for each additional one, so 1,000 images cost $14 to $17 across all three sizes in our table. Resolution barely matters at that rate. BFL's per-image models are simpler: FLUX 1.1 [pro] is $0.04 per image and FLUX 3 Image is $0.048 at 1K, rising to $0.607 at 4K.

That 4K FLUX 3 price deserves a flag. At $607 per 1,000 images, it is the most expensive cell on any rate card in this post. If you need 4K, compare it against Nano Banana 2.1 at $113 per 1,000 before you commit.

Commercial use and indemnity compared across image APIs

All three vendors let you use generated images commercially. They differ sharply on what happens when a third party says one of those images infringes their rights. That difference never shows up on a rate card, and in my view it is worth more than any price gap in the table above.

Output terms for API customers, as read on 10 November 2026
TermOpenAI APIGoogle Gemini APIBlack Forest Labs API
Output ownershipNot verified for this post. OpenAI's terms page blocked automated access."Google won't claim ownership" of generated content. You are responsible for its use."You own all right, title, and interest in and to Output."
Vendor IP indemnity on outputYes. Covers third-party claims that your use or distribution of Output infringes IP.Cloud terms cover unmodified output from listed, paid services. We could not confirm which image models are listed.None found. You indemnify BFL for claims arising from your use of Output.
Named exclusionsNot covered if you knew or should have known the Output was likely to infringe.Knew or should have known, bypassed filters, use after notice, trademark claims.Not applicable. BFL disclaims non-infringement warranties.
Liability cap or training rightsNot verified for this post.Not verified for this post.Cap of the greater of $100 or six months of fees. BFL takes a perpetual licence to inputs and outputs, including for training.

OpenAI and Google: indemnity with conditions

OpenAI's Service Terms, as captured on 4 October 2026 by the terms-monitoring service ConductAtlas, extend OpenAI's indemnity to API customers for "any third party claim that Customer's use or distribution of Output infringes a third party's intellectual property right." It does not apply where you knew or should have known the output was likely to infringe. We could not open OpenAI's own terms page, so treat this row as secondary and check the live text.

Google's Cloud Service Specific Terms cover unmodified generated output from services named on a separate indemnified-services list, and only where the service is not free. The exclusions include output you knew was likely infringing, disabled filters, use after notice of a claim, and trademark claims. The Gemini API's own additional terms contain no indemnity at all, and the list page returned no content when we fetched it. Ask your Google account team, in writing, whether Nano Banana 2.1 is on it.

Black Forest Labs: you carry the risk

BFL's developer terms make you indemnify BFL for claims arising from your use or distribution of Output. The same terms disclaim any warranty of non-infringement and cap BFL's total liability at the greater of $100 or what you paid in the prior six months. They also bar you from using outputs to train other AI models.

That is a defensible position for a model vendor, and it is common. It does mean the $30 per 1,000 headline on FLUX.2 [pro] is not like-for-like with OpenAI's $53 at medium. Part of what the higher price buys is a contractual promise BFL does not make. For the wider legal picture, see how AI copyright indemnity clauses compare across vendors and the Disney and Midjourney case on output copyright.

Where this comparison is weakest

Every number in the table is list price for output only. That framing is fair for a buyer comparing vendors, and it is also the place where a cost model most often goes wrong.

Cost per image is not cost per usable image. The table prices a generation, not a keeper. If one model needs four attempts to produce an image your brand team accepts and another needs one, the cheaper rate card loses. No vendor publishes acceptance rates, and nothing in this post measures quality. A $14 FLUX.2 [klein] image you regenerate five times costs more than a $53 gpt-image-2 image you keep.

The mappings are approximate. Google prices by tier, so its 1K landscape image is not the same pixel count as OpenAI's 1536 by 1024. OpenAI's gpt-image-2 per-image table sits under a "historical output costs" heading and was read through a page summary, not a static table we could copy. OpenAI publishes no per-image cost for its newest models at all. Each of these gaps can move a cell by a meaningful amount.

The terms column is the thinnest evidence here. We could read BFL's and Google's terms directly. OpenAI's terms page refused automated access, so its indemnity wording comes from a third-party capture dated 4 October 2026, and Google's indemnified-services list would not load. A lawyer with a browser will do better than this table in ten minutes, and you should ask one before relying on any indemnity.

Frequently asked questions

How much does it cost to generate 1,000 images with an AI image API?

On official list prices in November 2026, 1,000 square images at about 1 megapixel cost from $5 to $211. FLUX.2 [pro] costs $30, Gemini Nano Banana 2.1 costs $33.60 or $16.80 on batch, and gpt-image-2 costs $6, $53 or $211 at low, medium and high quality. Quality setting moves the bill more than vendor choice, so fix your quality default before you compare vendors.

What is GPT Image pricing per image?

OpenAI's published table for gpt-image-2 prices a 1024 by 1024 image at $0.006 on low, $0.053 on medium and $0.211 on high quality. Underneath, OpenAI bills image output at $30 per million tokens. gpt-image-2 is deprecated and leaves the API on 10 February 2027, and its replacements, gpt-image-2.6 Sunburst and Flare, use the same token rate without a published per-image table.

How much does Imagen cost on the Gemini API?

Imagen 4 is no longer sold on the Gemini API. Google shut down all three Imagen 4 model IDs on 17 August 2026 and names Gemini Nano Banana 2.1 as the replacement. Nano Banana 2.1 costs $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image, and the Batch API halves each of those prices for jobs that can wait.

How is Flux API cost calculated?

FLUX.2 models are billed per megapixel, with a higher rate for the first megapixel and a lower rate for each additional one. Black Forest Labs defines a megapixel as 1,048,576 pixels and rounds up. FLUX.2 [pro] costs $0.03 for a 1024 by 1024 image but $0.045 for a 1080 by 1080 image, because the larger square rounds up to 2 megapixels.

Which AI image API offers copyright indemnity?

OpenAI's API terms include an indemnity for third-party IP claims about output, with exclusions such as knowing infringement. Google's Cloud terms indemnify unmodified output from listed paid services, though the list should be checked for your image model. Black Forest Labs offers no output indemnity and instead requires developers to indemnify it, with its own liability capped at the greater of $100 or six months of fees.

Is batch image generation cheaper than real-time?

Yes, on two of the three vendors. Google's Batch API cuts Gemini image prices by 50%, taking Nano Banana 2.1 from $33.60 to $16.80 per 1,000 1K images. OpenAI halves its gpt-image-2 token rates on batch jobs but publishes no per-image batch prices. Black Forest Labs lists no batch discount on its pricing page.

How to pick an image API this week

Run one test, not a research project. Take 20 real prompts from last month's campaign and generate each one on FLUX.2 [pro] at 1024 by 1024, Nano Banana 2.1 at 1K on batch, and gpt-image-2.6 Flare at medium. Count how many images per model your brand lead would actually ship, then divide each vendor's cost by that number.

Then send one email to whichever vendor wins on cost per keeper. Ask whether its output indemnity covers that exact model, at your plan, and get the answer in writing. If the cheapest option cannot answer yes, price the risk before you price the images. For how credit-based design tools wrap these same models, see the breakdown of AI design tool credit pricing.

References

  1. OpenAI, API pricing, retrieved 10 November 2026. Used for image token rates and batch rates.
  2. OpenAI, Image generation guide, retrieved 10 November 2026. Used for gpt-image-2 per-image prices; gpt-image-1.5 and gpt-image-1-mini figures from their model pages.
  3. OpenAI, Deprecations, retrieved 10 November 2026. Used for gpt-image-1, 1.5, mini and 2 retirement dates.
  4. Google AI Studio, Gemini API pricing and Gemini API deprecations, both retrieved 10 November 2026. Used for Nano Banana 2.1, Gemini 3.1 Flash Image and Gemini 3 Pro Image prices, and for the Imagen 4 shutdown date and replacement. Vertex AI status from the Imagen 4 model page on docs.cloud.google.com.
  5. Black Forest Labs, Pricing, retrieved 10 November 2026. Used for all FLUX prices and the megapixel definition.
  6. Black Forest Labs, Developer Terms of Service, revised 4 August 2026, and FLUX API Service Terms, revised 4 August 2026. Used for ownership, indemnity, liability cap and training licence.
  7. Google, Gemini API Additional Terms of Service, effective 23 March 2026, and Google Cloud Service Specific Terms. Used for output ownership and indemnity conditions.
  8. ConductAtlas, OpenAI Service Terms: API output indemnification, capture of 4 October 2026. Secondary source for OpenAI's indemnity wording.

Weakest point in the source base: OpenAI's indemnity wording is a third-party capture because OpenAI's terms page blocked automated access, and Google's indemnified-services list did not load. All prices were last verified on 10 November 2026 and change often; re-check before February 2027, when gpt-image-2 leaves the API.

AV
Aryan Vatsa
Writes for Zan Digital about AI product economics, API pricing and what vendor rate cards actually cost a buyer.

Related reading