GPT Image 2 API pricing per image: 1K, 2K and 4K rates, monthly cost, and OpenAI's token math

GPT Image 2 costs $0.03 per image at 1K, $0.05 at 2K and $0.08 at 4K on UnificAlly, as a flat price per finished image rather than OpenAI's per-token meter.

That number is what most people searching for GPT Image 2 want, so the rest of this guide is the math around it: what 100, 1,000 and 10,000 images a month cost at each resolution, how the flat rate lines up against the token prices on OpenAI's own pricing page, the exact request the openai/gpt-image-2 route accepts, and where GPT Image 2 stands now that GPT Image 2.5 is out. The playground on that page is the fastest way to try it first.

TL;DR: GPT Image 2 (openai/gpt-image-2) accepts a prompt, a 1K, 2K or 4K resolution, an optional 1:1, 3:2, 2:3 or 16:9 aspect ratio, and up to 16 reference images. UnificAlly bills $0.03 / $0.05 / $0.08 per image, so 1,000 images at 1K cost $30. OpenAI's own API bills the same model per token. On Arena it sits at #3 on both image boards, behind only the two GPT Image 2.5 models. Submit one async task, then poll its task_id.

What is GPT Image 2?

GPT Image 2 is OpenAI's image generation and editing model, released on April 21, 2026 alongside ChatGPT Images 2.0 as the successor to GPT Image 1 and GPT Image 1.5. OpenAI's model page lists generation, editing and inpainting, with text and image input and image output; the default API snapshot is gpt-image-2-2026-04-21. UnificAlly exposes it through one task route with a prompt, a resolution, an aspect ratio and optional reference images.

The model supports:

• Photorealism and text rendering. OpenAI's launch post leads with layout control and multilingual text, with Japanese, Arabic, Korean, Devanagari and Cyrillic in its own examples. The 1K test below checks exact label copy in a commercial product scene. Clean text also makes it a practical pick for UI mockups and interface screenshots where labels have to read correctly.
• Reference-guided generation. Up to 16 reference images per call, each up to 100 MB.
• Simple integration. The same async generate-and-poll pattern as the rest of the OpenAI image family.

What's new in GPT Image 2

Two things stand out against GPT Image 1.5:

1. Text that reads. Multi-line headlines, signage and captions render correctly across scripts, which is the upgrade OpenAI built the launch around.
2. A flat price per resolution on UnificAlly. OpenAI's API still bills GPT Image 2 per token, with a quality setting (low, medium, high) that changes how many output tokens an image uses. The UnificAlly route replaces that with a resolution field: 1K costs $0.03, 2K $0.05 and 4K $0.08, so the request tells you the cost before it runs.

If you are still on gpt-image-1, OpenAI's deprecations page lists its shutdown for October 23, 2026; gpt-image-1-mini and gpt-image-1.5 follow on December 1, 2026. OpenAI names the GPT Image 2.5 models as the replacement for all three. GPT Image 2 itself is not on the deprecation list.

What you can do with GPT Image 2

Inputs

• Prompt. Required text describing the image.
• resolution. 1K, 2K or 4K; defaults to 1K. Drives both output size and per-image price.
• aspect_ratio. 1:1, 3:2, 2:3 or 16:9; defaults to 1:1.
• image_urls. Up to 16 optional reference images, each up to 100 MB.
• nsfw_check. Boolean, defaults to true; screens the prompt and any input images before generation.

Output

• A single image at the chosen resolution and aspect ratio.
• Returned via task-based async polling, same shape as other OpenAI image variants.

Three resolution variants

GPT Image 2 strengths showcase

We generated the output below through the UnificAlly API with openai/gpt-image-2 on August 13, 2026. Settings: 1K resolution, 16:9 aspect ratio, no reference images. The completed task was billed at the 1K rate.

Text rendering

Test prompt

Output

"Editorial product photograph for a fictional specialty coffee brand. A matte black coffee bag stands upright on a pale stone counter in soft morning window light. The front label must read exactly, on separate lines: NORTHLINE / NIGHT ROAST / 340 g. Add a small cream price card beside it reading exactly: $18. Keep every word sharp and correctly spelled. Realistic paper texture, restrained charcoal and cream palette, natural shadows, clean commercial composition, no extra text, no logos beyond the requested fictional label."

This render reproduces all four requested text elements correctly: NORTHLINE, NIGHT ROAST, 340 g, and $18. The label hierarchy stays readable against the textured paper, and the price card remains distinct without adding stray copy. This is one 1K generation, so it does not establish repeatability or test reference-image editing, 2K, or 4K output.

GPT Image 2 benchmarks: Arena rank

Arena's human-vote leaderboards are the closest thing to a neutral quality score for image models. GPT Image 2 holds third place on both boards, and the only models above it are OpenAI's own GPT Image 2.5 pair.

Arena lists the model as "GPT-Image-2 (Medium)", the medium quality setting on OpenAI's API. The two GPT Image 2.5 rows are still marked preliminary with far fewer votes, so the gap at the top can move; GPT Image 2's own scores rest on hundreds of thousands of votes and have been stable for months. Our read: for the price, GPT Image 2 is still the best-ranked image model you can call at a flat $0.03 per 1K image, and the upgrade case for 2.5 is about edit precision rather than raw ranking.

GPT Image 2 API pricing per image

These rates come from UnificAlly's live price list and refresh on every deploy. Only completed generations are billed; a task that fails costs nothing. The pricing page lists every model side by side.

Run GPT Image 2 at 1K, 2K or 4K in the UnificAlly playground.

GPT Image 2 cost per month at 100, 1,000 and 10,000 images

Because the rate is flat, monthly cost is count times rate with nothing to estimate. The table multiplies the live per-image rates above by three common volumes.

Most production mixes are not all one size. Draft at 1K, keep the prompt, then rerun only the keepers at 2K or 4K: a team that generates 1,000 drafts and promotes 100 of them to 4K pays $30 plus $8, well under the cost of 1,000 images at 4K. There is no subscription and the balance does not expire, so a slow month costs only what you generate.

GPT Image 2 vs OpenAI's token pricing

On OpenAI's own API, GPT Image 2 has no per-image price. The OpenAI pricing page bills it per million tokens, and how many output tokens one image uses depends on its size and the quality setting, so the official cost of a 1,000-image month is an estimate until the bill arrives. UnificAlly does not bill per token: a 1K generation is $0.03 whether you send zero references or sixteen, and the prompt, the reference images and the output are all inside that one rate. If you need to predict spend before a batch runs, the flat rate is the simpler model; if you already run OpenAI batch jobs with short prompts and no references, price both for your real token counts before moving.

How to call GPT Image 2

The UnificAlly GPT Image 2 documentation defines the current request fields. The API is async: POST a generation, then poll until the image URL is ready.

const API = 'https://api.unifically.com';
const headers = {
  Authorization: `Bearer ${process.env.UNIFICALLY_API_KEY}`,
  'Content-Type': 'application/json',
};

const start = await fetch(`${API}/v1/tasks`, {
  method: 'POST',
  headers,
  body: JSON.stringify({
    model: 'openai/gpt-image-2',
    input: {
      prompt: 'A photorealistic packaging mockup of a matte black cosmetics bottle on a marble surface, soft side lighting, brand wordmark in clean serif type',
      resolution: '4K',
      aspect_ratio: '16:9',
      image_urls: ['https://example.com/brand-reference.jpg'],
    },
  }),
}).then((r) => r.json());

while (true) {
  await new Promise((r) => setTimeout(r, 2000));
  const task = await fetch(`${API}/v1/tasks/${start.data.task_id}`, { headers }).then((r) => r.json());
  if (task.data.status === 'completed') {
    console.log(task.data.output.image_url);
    break;
  }
  if (task.data.status === 'failed') throw new Error(task.data.error_message);
}

To draft at the lowest rate, set resolution: '1K'. A 1K generation costs $0.03. Treat it as a draft rather than a guaranteed preview of the 4K result because each generation can vary.

Open the GPT Image 2 API playground and run this request.

GPT Image 2 vs GPT Image 2.5

OpenAI released GPT Image 2.5 on September 8, 2026 as two models, gpt-image-2.5-flare and gpt-image-2.5-sunburst, and its deprecations page now names them, not GPT Image 2, as the replacement for GPT Image 1, 1 mini and 1.5. Both are callable on UnificAlly from the GPT Image 2.5 API page with the same request shape as GPT Image 2.

The decision is narrower than the version numbers suggest. For plain text-to-image, Flare costs the same per resolution as GPT Image 2 on UnificAlly and ranks above it, so it is the better default for new work. For reference-heavy editing, GPT Image 2 includes all 16 references in its flat rate while both 2.5 models add a per-reference surcharge (listed on their page), so a product-catalog pipeline that sends six references per call stays cheaper on GPT Image 2. Sunburst earns its higher rate only on final assets that go through long edit chains, where holding a subject steady matters more than the extra cents.

GPT Image 2 vs Nano Banana 2

Google replaced Nano Banana 2 with Nano Banana 2.1 on October 6, 2026 and marked the older model deprecated the same day, so the live comparison is now GPT Image 2 against 2.1. Both run at 1K, 2K and 4K on UnificAlly; their prices, aspect ratios and reference limits differ.

Nano Banana 2.1 is the cheaper call at every size and the only one of the two with a vertical 9:16 ratio and an auto option that follows the reference. GPT Image 2 ranks higher on both Arena boards, takes more references, and is the one to test first for posters, packaging and anything where the text has to read. Run the same prompt and references through both playgrounds and compare text accuracy, composition and reference consistency before a batch; the Nano Banana 2.1 vs Nano Banana 2 pricing post covers what changed on Google's side.

Migrating off gpt-image-1

OpenAI's deprecations page lists gpt-image-1 shutdown on October 23, 2026, with gpt-image-1-mini and gpt-image-1.5 shutting down on December 1, 2026. OpenAI's named replacement for all three is gpt-image-2.5-sunburst or gpt-image-2.5-flare. DALL·E 2 and DALL·E 3 already shut down on May 12, 2026, with GPT Image 2 among the models OpenAI listed as their replacement. If your codebase still references any of them, two practical paths:

• Move gpt-image-1 work before October 23, 2026 to GPT Image 2 or GPT Image 2.5. On UnificAlly both take the same prompt, resolution, aspect_ratio and image_urls fields, so the migration is a model-name change plus a test run.
• Move gpt-image-1.5 and gpt-image-1-mini work before December 1, 2026. Same request shape, same two targets.

Test the same production prompts on the new model because output style can change even when the request fields look similar.

Things to know about GPT Image 2

• Defaulting to 4K when 2K is the actual delivery target. A 16:9 web hero rarely needs 4K. 4K costs $0.08 per image against $0.05 at 2K, so match the resolution to the destination.
• Sending references that do not serve a clear purpose. GPT Image 2 supports up to 16 references at 100 MB each. Start with the images needed to define the subject, layout, or style, then add more only when they supply missing information.
• Not migrating off gpt-image-1 before October 23, 2026. OpenAI shuts the route down on that date. Calls to gpt-image-1 will start returning errors after deprecation.
• Mixing GPT Image 2 with another model at random. If output style matters, use one model per asset class. Switching mid-workflow can introduce visual drift.
• Sending a quality field to the UnificAlly route. The UnificAlly wrapper exposes resolution instead. Use 1K, 2K, or 4K.

GPT Image 2 FAQ

Is gpt-image-2 the same as GPT Image 2.0?

Yes, they are the same model written differently, and developers also search the API route IDs directly (gpt-image-2, gpt-image-2-4k for the 4K tier). The API price per image is identical however you write it: the same rate card above, with 4K priced on its own row.

What is GPT Image 2?

GPT Image 2 is OpenAI's image generation and editing model, released April 21, 2026. On UnificAlly it accepts a prompt, a resolution (1K, 2K, or 4K), an optional aspect ratio (1:1, 3:2, 2:3, or 16:9), and up to 16 reference images. Output is a single image returned through async task polling.

How much does GPT Image 2 cost?

On UnificAlly, GPT Image 2 costs $0.03 at 1K, $0.05 at 2K, and $0.08 at 4K. Billing is per generated image against the openai/gpt-image-2 price key, and only completed generations are charged. The pricing page lists every model's rates.

What do 1,000 GPT Image 2 images cost?

At the current UnificAlly rates, 1,000 images cost $30 at 1K, $50 at 2K, or $80 at 4K. The rate is flat per image, so a mixed batch is the sum of each resolution's count times its rate.

How much does the OpenAI image generation API cost?

OpenAI bills GPT Image 2 per million tokens, with separate rates for image output, image input, text input and cached input, and the per-image cost depends on the size and quality setting. UnificAlly bills the same model at a flat per-image rate by resolution: $0.03 at 1K, $0.05 at 2K, and $0.08 at 4K.

Is gpt-image-1 still available?

gpt-image-1 is scheduled for shutdown on October 23, 2026, and gpt-image-1-mini and gpt-image-1.5 on December 1, 2026. OpenAI names the GPT Image 2.5 models as their replacement. GPT Image 2 is not deprecated, and on UnificAlly both GPT Image 2 and GPT Image 2.5 accept the same request fields.

When was GPT Image 2 released?

GPT Image 2 was released on April 21, 2026 as part of the ChatGPT Images 2.0 launch, with the API snapshot gpt-image-2-2026-04-21. Compared to GPT Image 1.5 it improved text rendering and raised the reference-image limit to 16 per call.

Is GPT Image 2 better than GPT Image 2.5?

No on the leaderboards: as of October 7, 2026 GPT Image 2.5 Sunburst and Flare rank #1 and #2 on Arena's text-to-image and image-edit boards, with GPT Image 2 at #3 on both. On UnificAlly, GPT Image 2.5 Flare costs the same per resolution as GPT Image 2 but both 2.5 models add a per-reference-image surcharge, so GPT Image 2 stays the cheaper choice for reference-heavy editing.

How many reference images can I send?

GPT Image 2 supports up to 16 reference images in one call, each up to 100 MB. Start with the smallest set that clearly defines the subject or style, then add references only when they supply missing information.

How does GPT Image 2 compare to Nano Banana 2?

Nano Banana 2 was deprecated by Google on October 6, 2026 in favor of Nano Banana 2.1, which costs less than GPT Image 2 at every resolution on UnificAlly and adds a 9:16 ratio and an auto option. GPT Image 2 ranks higher on both Arena boards (#3 against #6) and takes 16 references to Nano Banana 2.1's 10, so it is the first test for text-heavy layouts.

If you need OpenAI image generation with predictable per-image pricing, open GPT Image 2 and start with a $0.03 1K draft, then move the final prompt to 2K or 4K only when the destination requires it.

Related reading on GPT Image 2

• Run GPT Image 2: live playground with 1K, 2K, and 4K resolution controls.
• GPT Image 2.5 API: Flare and Sunburst, the models OpenAI now names as the GPT Image 1 and 1.5 replacement.
• GPT Image 1 vs 1.5 vs 2: the migration guide for older GPT Image pipelines.
• GPT Image 2 vs Nano Banana: the longer head-to-head with benchmarks.
• Nano Banana 2.1 vs Nano Banana 2 and the Nano Banana 2.1 API: Google's current image model and what changed.
• Nano Banana Pro: Gemini 3 Pro Image at 1K, 2K or 4K.
• Flux.2: Black Forest Labs option with Flex / Pro / Max variants.