How Much To Run AI
Hosted API

GPT Image 2.6: Flare vs Sunburst, API Costs, GIFs and SVG

GPT Image 2.6 is OpenAI’s image generation and editing release, available in the API as gpt-image-2.6-flare and gpt-image-2.6-sunburst. Choose Flare for fast iteration or Sunburst for detailed visual work. Both support complexity controls, markup-guided edits, GIFs and SVG through the Image API. API cost depends on the text and image tokens in each request, so there is no fixed price per image.

Content reviewed:

GPT Image 2.6 token rates of $5 per million text input, $8 per million image input and $30 per million image output, with one measured 1024 × 1024 request at $0.044015 and one 512 × 512 GIF at $0.055880.
Original diagram of the published rates; the two dollar figures come from one published test each, not from an average price.

Use this model

API docs and console

Choose a workflow

  • Create everyday campaign images or explore several visual directions with Flare.
  • Work on detailed compositions and iterative edits with Sunburst.
  • Export an animated GIF or a static SVG from a generation request.

Choose Flare for iteration or Sunburst for detail

Both API models generate and edit images from text and image inputs. OpenAI positions Flare for fast iteration and everyday image work, while Sunburst is for detailed visual work. OpenAI describes the Sunburst model as useful when a project needs more precision and consistency across edits at medium or high quality. Compare both on the same prompt and reference images before choosing a default.

For GPT Image 2.6, use the Image API generation and edit endpoints. OpenAI also provides image generation inside ChatGPT and Codex; those product plans do not show the same per-request API token bill. API access may require organization verification, and rate limits depend on your organization’s usage tier.

Both models expose a complexity setting (low, medium or high) alongside rendering quality. Complexity controls how much planning happens before rendering and has no separate fee, but a more complex request can take longer. The default complexity is high. Start lower for a focused change and compare actual usage before choosing a default for a repeated workflow.

Estimate API cost from token usage

As checked November 6, 2026, both 2.6 models list the same Standard API rates: $5 per million text input tokens, $1.25 per million cached text input tokens, $8 per million image input tokens, $2 per million cached image input tokens and $30 per million image output tokens. Token usage can differ between models and settings, so the same rates can produce different per-image charges.

A November 4 API test by Curtis Pyke at Kingy reported one 1024 × 1024 medium-quality Sunburst product image using 109 text input tokens and 1,449 image output tokens. At the listed rates, that request calculated to $0.000545 + $0.043470 = $0.044015, excluding any account-specific adjustments. The author ran each prompt once and calculated costs from returned usage.

For a budget, add prompt text, any input-image tokens, generated image tokens and every revision. Read the returned usage for your own settings, then multiply each token count by its current rate. Retries and edits that produce another output add another image charge.

Token typeStandard / 1MCached input / 1M
Text input$5.00$1.25
Image input$8.00$2.00
Image output$30.00—

Generate and save your first image

Create an OpenAI API key, check that your organization can use GPT Image models, set OPENAI_API_KEY in your environment, and install the OpenAI Python package. The example requests one medium-quality square image with Flare. Swap in gpt-image-2.6-sunburst when detailed visual work matters more. The Images API returns image data as base64; decode it to save the file.

After the first request, keep the model ID, size, quality, prompt, response usage and output together. Inspect exact text and small details at the final delivery size. If you need another version, count that edit or reroll as another generation in your working budget.

Python SDK: one square image through the Images API
import base64
from openai import OpenAI

client = OpenAI()
result = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="A clean editorial illustration of a blue ceramic mug on a pale wood desk, soft window light, no text",
    size="1024x1024",
    quality="medium",
)
with open("image.png", "wb") as image:
    image.write(base64.b64decode(result.data[0].b64_json))

Edit an image or generate an animated GIF

Use the Images API edit endpoint with your source image, an explicit 2.6 model ID and a prompt for the change. Markup-guided edits let you attach a marked-up layer that points to the region to change; OpenAI lists no extra fee for the markup itself. The request still bills its input and output tokens. Set the model explicitly on every edit so the request uses the version you intend.

For native GIF output, use the generation or edit endpoint with output_format set to gif. GIF output requires preview access for the selected model; request it if it is not enabled. OpenAI’s current API reference allows 2–16 frames (default 9) and 20–2,000 milliseconds per frame (default 200, rounded to 10 ms). The frame size and total pixel budget must fit the selected model’s limits. GIF output does not support streaming, partial images or custom output compression.

In Kingy’s single November 4 Sunburst API test, a 512 × 512, two-second GIF used 10 frames at 200 ms and calculated to $0.055880. The evaluator found the motion uneven and the return to the first pose noticeable. Inspect the transition back to the first frame before accepting a loop.

Export static vector artwork as SVG

Set output_format to svg on an Images API generation request when you need editable vector shapes, such as an icon or simple illustration. SVG generation requires n=1; edits cannot return SVG. The result is static vector content without scripts, external resources or embedded raster images.

SVG billing includes the image model’s tokens for its reference image and the selected SVG authoring model’s tokens for writing the vector document. OpenAI does not give a fixed SVG surcharge: the amount depends on usage from both models. Keep the returned usage and authoring-model choice with your estimate.

Before you use it

  • API token pricing is separate from ChatGPT or Codex plan access; check organization verification and account rate limits before building around the API.
  • Per-image cost varies with token usage, image size, quality, input references and edits. The cited community examples are individual runs, not average prices.
  • SVG is generation-only and bills both its image reference and authoring-model usage. GIF output has frame and total-pixel limits and requires preview access for the selected model.

How to get started

Start with one representative brief at a fixed size and quality. Compare Flare and Sunburst only where both could fit, record each response’s token usage and review the full-size image or GIF loop before accepting it.

Sources

Sources (9)

Calculators