gpt-image-2 API: price per image, limits and code

gpt-image-2 price per image

Text and image input tokens are billed on top of the per-image output cost. Square images are typically fastest to generate, and a larger non-square size can sometimes use fewer output tokens than a smaller or square one at the same quality.

The table below shows the output cost per gpt-image-2 image by quality and size.

Token prices, Standard and Batch

Per-token rates decide the real cost of image edits and long prompts.

• Cached input rates only apply to images generated with the Responses API, so Images API calls pay the full input rate.
• gpt-image-2.5-sunburst and gpt-image-2.5-flare cost the same per token as gpt-image-2 and add xhigh and max quality settings.

The table below lists gpt-image-2 prices per 1M tokens on the Standard and Batch tiers; see OpenAI API pricing for every other model.

Endpoints, sizes and parameters

• Endpoints: gpt-image-2 (default snapshot gpt-image-2-2026-04-21) works on POST /v1/images/generations, POST /v1/images/edits, and the /v1/batch Batch API, while Chat Completions (/v1/chat/completions) and the Responses API (/v1/responses) are listed as not supported.
• Size: Pass auto (the default), popular presets like 1024x1024, 1536x1024, 1024x1536, 2048x2048, 2048x1152, 3840x2160, or 2160x3840, or any custom dimensions up to a 3840 px longest edge where both edges are multiples of 16 px, the aspect ratio is at most 3:1, and total pixels sit between 655,360 and 8,294,400.
• Quality: Set low, medium, high, or auto (the default), with low suggested for quick drafts.
• n: Generates several images in one request, and defaults to returning one image.
• Output: The API returns base64-encoded image data (b64_json) as png by default, or jpeg and webp with output_compression from 0 to 100 (jpeg is faster than png).
• Fidelity: Leave input_fidelity out, because gpt-image-2 processes every image input at high fidelity automatically.

Rate limits and latency

Complex prompts can take up to 2 minutes to process, so set long client timeouts. OpenAI may also ask for API Organization Verification before you can use GPT Image models.

The table below shows gpt-image-2 rate limits by OpenAI usage tier.

Is gpt-image-2 free?

OpenAI's API has no free tier for gpt-image-2. Pay-as-you-go API accounts prepay at least $5, requests fail with a billing error once credits run out, and the model page lists no rate limits for the Free usage tier. The ChatGPT Pro and Business API credits benefit ($100 to $500 a month on Pro) does cover the Images API; see our guide to free OpenAI API options.

In Codex on ChatGPT plans, image generation isn't available on the Free plan at all. On paid plans, it uses your included limits about 3 to 5x faster on average than similar turns without image generation, depending on quality and size.

Generate gpt-image-2 images with your ChatGPT plan

Plan2API's POST /v1/images/generations and POST /v1/images/edits endpoints render gpt-image-2 on your own connected ChatGPT (Codex) subscription, so a plan with Codex such as Plus or Pro does the work and Plan2API never bills per token. You connect the plan once (see ChatGPT subscription API), but the image endpoints are available on Plan2API Pro ($12 a month) and Scale only, and calling them on the Free plan returns HTTP 403 with type plan_limit (see pricing for plan details).

There are a few differences from OpenAI's direct API when routing through your subscription: size sets the aspect ratio only (from 1:3 to 3:1, about 1.57 megapixels per image), output is PNG as b64_json only, n is 1 to 4, there is no streaming and no masks, and edits take up to 5 source images. A render takes about 15 to 45 seconds, and up to about 110 seconds at high quality, so set your client timeouts above 2 minutes.

curl https://api.plan2api.com/v1/images/generations \
  -H "Authorization: Bearer $PLAN2API_KEY" \
  -H "Content-Type: application/json" \
  --max-time 180 \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A watercolor fox reading a map, soft morning light",
    "n": 1,
    "size": "1536x1024",
    "quality": "high",
    "response_format": "b64_json"
  }' | jq -r '.data[0].b64_json' | base64 --decode > fox.png