GPT Image 2 API: Specs, Pricing, and How to Call It (2026)

GPT Image 2 is OpenAI's current image generation model. It has the cleanest detail and tightest prompt adherence in the GPT Image family, with the simplest request shape OpenAI has released on this surface to date. On Unifically it comes in three resolution variants ($0.03 / $0.05 / $0.06 per image for 1K / 2K / 4K) and accepts up to 16 reference images per call. This guide covers what GPT Image 2 does, how it is priced, how to call it, and the most common mistakes teams make on day one.

TL;DR: GPT Image 2 (openai/gpt-image-2) is OpenAI's latest image route. It accepts a prompt, a resolution (1K, 2K, or 4K), an optional aspect ratio (1:1, 2:3, 3:2), and up to 16 reference images at 100 MB each. Per-image pricing on Unifically is $0.03 / $0.05 / $0.06. Same async generate-and-poll pattern as every other Unifically model. Live rates: pricing page.

What is GPT Image 2?

GPT Image 2 is OpenAI's newest image generation model. It accepts a prompt, an output resolution (1K, 2K, or 4K), an optional aspect ratio, and optional reference images for reference-guided generation. Compared to earlier GPT Image variants it has a simpler request shape: a single resolution control replaces the older "low / medium / high quality" knob, so you pick output size and pay accordingly.

The model is tuned for:

Photorealism and text legibility. Cleaner reflections, materials, and in-image type than earlier GPT Image generations.
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 changed in early 2026 that made GPT Image 2 the default starting point on the OpenAI side:

Resolution variants replaced quality variants. GPT Image 2 collapses the old "low / medium / high quality" knob into an explicit output resolution, which lines up with how teams actually budget image spend (drafts at 1K, finals at 4K).
Per-variant shape matches Google Nano Banana 2. Both list 1K / 2K / 4K at $0.03 / $0.05 / $0.06, so you can A/B test OpenAI vs Google output on the same workflow without juggling two pricing models.

If you are still on gpt-image-1, note that OpenAI's deprecations page lists gpt-image-1 shutdown for October 23, 2026. Production workloads on the older model need a migration plan. GPT Image 2 (or the more conservative GPT Image 1.5) is the destination.

What you can do with GPT Image 2

Inputs

Prompt. Required text describing the image.
resolution. 1k, 2k, or 4k. Drives both output size and per-image price.
aspect_ratio. 1:1, 2:3, or 3:2.
image_urls. Up to 16 optional reference images, each up to 100 MB.

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

Variant	Best for	Per-image price
GPT Image 2 1K	Drafts, social tiles, large-volume batches	$0.03
GPT Image 2 2K	Web hero art, presentation graphics	$0.05
GPT Image 2 4K	Hero campaign images, packaging, print-ready	$0.06

GPT Image 2 pricing and how it compares

Source	Variant	Per-image price
Unifically	GPT Image 2 1K	$0.03
Unifically	GPT Image 2 2K	$0.05
Unifically	GPT Image 2 4K	$0.06
Unifically	Nano Banana 2 (1K / 2K / 4K)	$0.03 / $0.05 / $0.06
Unifically	Nano Banana Pro (1K / 2K)	$0.06
Unifically	Flux.2 Klein 4B	~$0.0098 per MP
OpenAI direct	gpt-image-2 (1024×1024, low / medium / high)	~$0.006 / $0.053 / $0.211
OpenAI direct	gpt-image-1.5 (1024×1024, low / medium / high)	~$0.009 / $0.034 / $0.133

Source for OpenAI direct rates: published GPT Image 2 vs GPT Image 1.5 quality-variant pricing (1024×1024 reference). Unifically's resolution-priced variants map cleanly onto OpenAI's high-quality output rates while staying flat per-resolution, with no token-counting math required. Check the pricing page for current rates.

How to call GPT Image 2

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: '2:3',
      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 cheaply on the same prompt and references, set resolution: '1k'. The 1K variant costs $0.03 per image and reads as a faithful preview of the 4K final.

GPT Image 2 vs Nano Banana 2

The two models price identically ($0.03 / $0.05 / $0.06 across 1K / 2K / 4K) and accept similarly shaped inputs, so the right way to choose is on output character, not on cost.

GPT Image 2 runs OpenAI's image stack. Strong at photoreal materials, polished marketing visuals, and clean in-image typography. Up to 16 reference images at 100 MB each.
Nano Banana 2 runs Google's Gemini 2.5 Flash Image. Strong at character consistency across edits, ten aspect ratios from 1:1 through 21:9, and up to 20 reference images. SynthID watermarking is embedded automatically.

For brand-driven photoreal work where text legibility matters, GPT Image 2 is usually the better starting point. For character-led series and broad aspect-ratio coverage, Nano Banana 2 is the better starting point. The smart workflow on Unifically is to run the same prompt against both at the 1K variant ($0.03 each) and pick the model whose output wins your eye.

Migrating off `gpt-image-1`

OpenAI's deprecations page lists gpt-image-1 shutdown on October 23, 2026. If your codebase still references it, plan the move before then. Two practical migration paths:

Move to gpt-image-2. Newest model, simpler request shape, resolution variants. Best for new workloads or workloads where the latest model behaviour is acceptable.
Move to gpt-image-1.5. Stable production target, similar prompt behaviour to gpt-image-1, slightly cheaper at the medium / high quality bands. Best for low-risk migrations where output stability matters more than newness.

The Unifically wrappers for the deprecated variants (gpt-image-1, gpt-image-1.5) are hidden in the live pricing UI but the API is still callable for in-flight workloads.

Things to know

Defaulting to 4K when 2K is the actual delivery target. A 16:9 web hero rarely needs 4K. Match the variant to the destination, not the maximum.
Sending too many references. GPT Image 2 supports up to 16 references at 100 MB each. Three to five well-chosen references usually outperform sixteen mixed ones, since references compete for influence.
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 Nano Banana 2 at random. Both have the same per-variant shape ($0.03 / $0.05 / $0.06). If output style matters, lock to one model per asset class. Switching mid-workflow introduces visual drift.
Relying on legacy "quality" knobs that no longer exist. GPT Image 2 dropped the quality parameter from older variants. Set resolution instead.

Frequently asked questions

What is GPT Image 2?

GPT Image 2 is OpenAI's current image generation model. It accepts a prompt, a resolution (1K, 2K, or 4K), an optional aspect ratio (1:1, 2:3, or 3:2), and up to 16 reference images. Output is a single image returned via async task polling.

How much does GPT Image 2 cost?

On Unifically, GPT Image 2 prices per image by resolution variant: $0.03 at 1K, $0.05 at 2K, and $0.06 at 4K. There is no subscription. Billing is per generated image against the openai/gpt-image-2 price key. Check the pricing page for current rates.

Is `gpt-image-1` still available?

gpt-image-1 is OpenAI-deprecated and scheduled for shutdown on October 23, 2026. The route still works in-flight but is hidden in the Unifically pricing UI. Migrate to GPT Image 2 (or GPT Image 1.5 for stability) before that date.

How many reference images can I send?

GPT Image 2 supports up to 16 reference images in one call, each up to 100 MB. In practice, three to five well-chosen references give better results than sixteen mixed ones, since references compete for influence.

How does GPT Image 2 compare to Nano Banana 2?

Both have the same $0.03 / $0.05 / $0.06 per-variant price for 1K / 2K / 4K. GPT Image 2 runs OpenAI's image stack and tends to win on photoreal marketing visuals and text legibility. Nano Banana 2 runs Google's Gemini 2.5 Flash Image and tends to win on character consistency and broad aspect-ratio coverage.

GPT Image 2 model page: live playground with 1K / 2K / 4K toggle.
Nano Banana 2: same per-variant pricing, different model stack.
Nano Banana Pro: Gemini 3 Pro Image at 1K or 2K.
Flux.2: Black Forest Labs option with Flex / Pro / Max variants.
Suno and Veo 3.1: pair generated images with audio and video on the same account.

GPT Image 2 API: Specs, Pricing, and How to Call It (2026)

What is GPT Image 2?

What's new in GPT Image 2