GPT Image 2.6 API: Flare vs Sunburst, Cost per Image and Code

Updated 10 November 2026. OpenAI released GPT Image 2.6 Flare and Sunburst on 4 November; prices and limits below were checked on 10 November 2026.

GPT Image 2.6 reaches the API as two models, gpt-image-2.6-flare and gpt-image-2.6-sunburst, and they cost the same. OpenAI released both on 4 November 2026, alongside Images 2.6 in ChatGPT and Codex. Flare is the speed model: at low quality it can return an image in as little as six seconds. Sunburst is the detail model, with higher quality and more consistent edits at medium or high quality. Both add SVG export, animated GIF output, markup-guided edits and a new complexity setting.

As of 10 November 2026 both charge $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. OpenAI's own calculator puts a 1024x1024 image at about $0.025 at low quality and $0.053 at high, counting output only. One catch: low-quality drafts now cost about four times what they did on GPT Image 2.5. Below: Flare vs Sunburst, a cost table, Python code for images, SVG, GIFs and markup edits, and the limits to know.

Related: Pennsylvania AI Ad Disclosure Law: What to Label by December 21

GPT Image 2.6 at a glance

From OpenAI's Images 2.6 announcement, the image generation guide and the pricing page, checked on 10 November 2026:

Flare vs Sunburst: which should you use?

Because the token prices are identical, choose on speed and edit quality:

A practical pattern: draft with Flare at low quality, pick the winners, then make finals with Sunburst at medium or high. OpenAI's calculator gives one estimate for both models, but OpenAI also warns that token use can differ by model and quality, so read the usage field in each response.

How much does GPT Image 2.6 cost per image?

You pay per token: text input, plus image input when you edit, plus image output. Output is the main cost. These estimates come from the calculator in OpenAI's image generation guide; they exclude input tokens and streaming partial images, which add 100 output tokens each:

Is GPT Image 2.6 more expensive than 2.5?

Per token, no: GPT Image 2.5 lists the same rates. Per image, it depends on quality. For a 1024x1024 image, OpenAI's calculator and its historical price table give:

So if your workflow runs on cheap low-quality drafts, 2.6 costs about four times more per draft. At high quality, 2.6 matches 2.5 and costs about a quarter of gpt-image-2's high setting. There is also no Batch discount for 2.6; OpenAI's Batch table lists gpt-image-2, not the 2.6 models.

Worked example: 1,000 product images a month

Example with assumptions: a store makes three Flare drafts at low quality for each of 1,000 products, then one Sunburst final at high quality, all square. Drafts: 3,000 x $0.02469 = $74.07. Finals: 1,000 x $0.05268 = $52.68. Total: about $127 a month in output tokens. A 100-token prompt adds only $0.0005, but edits that send product photos add image input at $8 per 1M tokens. Our API costs guide shows how to track this per request.

How to use the GPT Image 2.6 API in Python

1. Create an API key and set OPENAI_API_KEY.
2. Check the developer console: OpenAI says you may need API Organization Verification before GPT Image models work.
3. Install the SDK with pip install openai.

import base64
from openai import OpenAI

client = OpenAI()

result = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="Studio photo of a matte black steel water bottle on a light gray background, soft shadow, no text",
    size="1024x1024",
    quality="low",
    output_format="jpeg",
    extra_body={"complexity": "low"},
)

with open("bottle-draft.jpg", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

print(result.usage)

extra_body sends fields your SDK version may not list yet, such as complexity. OpenAI says lower complexity suits focused edits, while higher complexity helps with dense text and complex layouts but can add latency. Adjusting it has no separate fee. JPEG returns faster than PNG; convert finals for the web with our image format guide.

The same request with curl:

curl -X POST "https://api.openai.com/v1/images/generations" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-image-2.6-sunburst", "prompt": "Flat lay of a leather wallet on white marble, soft daylight", "quality": "high", "size": "1536x1024"}' \
  | jq -r '.data[0].b64_json' | base64 --decode > wallet.png

How to export SVG with GPT Image 2.6

Set output_format to svg on the generations endpoint. Rules from OpenAI's guide: n must be 1; streaming, partial images and output_compression are not supported; the edit endpoint cannot return SVG or accept one as input. The API first makes a reference image, controlled by quality, then writes the SVG, so you pay the image rate for the reference plus the rate of the model that writes the SVG. The file is static: no scripts, external resources or embedded raster images.

import base64
from openai import OpenAI

client = OpenAI()

result = client.images.generate(
    model="gpt-image-2.6-sunburst",
    prompt="Simple flat logo for a bakery called Field and Flour: a wheat stalk inside a circle, two colors",
    output_format="svg",
    n=1,
    quality="medium",
)

with open("logo.svg", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

How to make an animated GIF with the API

Set output_format to gif on generations or edits; use edits to animate from a still reference. frame_count takes 2 to 16 frames (default 9) and frame_duration_ms takes 20 to 2,000 milliseconds, rounded to 10 (default 200). Nine frames at 200 ms make a 1.8-second loop. size sets each frame, and frames, size and n must fit the pixel budget. GIFs do not stream.

import base64
from openai import OpenAI

client = OpenAI()

result = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="A paper coffee cup with steam rising and the lid wobbling, flat illustration, smooth loop that ends where it starts",
    output_format="gif",
    extra_body={"frame_count": 9, "frame_duration_ms": 200},
)

with open("coffee-loop.gif", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

How to edit a product photo with markup

Markup is a transparent PNG layer where you circle or scribble on the area to change, then describe the change in the prompt. The layer must match the first image's width and height, each can hold up to 16,777,216 pixels, the layer file is limited to 50 MB, and you cannot combine markup with a mask. Markup edits use the normal token rates. This illustrative example follows the pattern in OpenAI's guide:

import base64
from pathlib import Path
from openai import OpenAI

client = OpenAI()

with open("sofa.png", "rb") as image, open("sofa-markup.png", "rb") as markup:
    result = client.images.edit(
        model="gpt-image-2.6-flare",
        image=("sofa.png", image, "image/png"),
        markup=("sofa-markup.png", markup, "image/png"),
        quality="medium",
        prompt=(
            "Change the cushion inside the red circle to mustard yellow linen. "
            "Keep the sofa, the room and the lighting unchanged. "
            "Remove the red circle from the finished photo."
        ),
    )

Path("sofa-yellow.png").write_bytes(base64.b64decode(result.data[0].b64_json))

OpenAI recommends low complexity for most edits and medium or high for scenes with many elements or constraints.

Limits and gotchas

• Sizes: both edges must be multiples of 16, the aspect ratio must sit between 1:3 and 3:1, no edge may pass 3,840 px, and total pixels must be 655,360 to 8,294,400. Sizes above 2560x1440 are experimental.
• Latency: complex prompts can take up to 2 minutes.
• Known weak spots: OpenAI lists precise text placement, recurring characters and tight layouts as areas where results can still slip.
• End users: send a hashed safety_identifier of 128 characters or fewer; the old user field is deprecated.
• Moderation: blocked requests return a moderation_blocked error with categories, so show users a clear retry message.

Is GPT Image 2.6 free in ChatGPT?

OpenAI says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers, on desktop, mobile and web. It did not publish per-plan limits for images or GIFs on the pages we read, so check the limits shown in your own account. API use is separate and billed per token, as above.

What can a store or marketing team build with it?

A hypothetical catalog pipeline: one step writes the listing with an AI product description generator, Flare drafts lifestyle images, a person picks one, and Sunburst renders the final. Our guides to bulk product descriptions from a CSV and the product description prompt cover the text side. Check marketplace rules first: Etsy's October policy update requires an unedited photo of the real item on most listings.

We build tools like these on fixed quotes, and you own 100% of the code. If you want a draft-to-final image pipeline costed for your catalog, send us the details, or browse more automation work.

GPT Image 2.6 FAQ

What is the difference between GPT Image 2.6 Flare and Sunburst?

Flare is built for speed: OpenAI says it can generate an image in as little as six seconds at low quality and improves layouts for posters and product imagery. Sunburst is built for detail, with higher quality and more consistent edits at medium or high quality. Both cost the same per token and support SVG, GIF, markup edits and the complexity setting.

How much does GPT Image 2.6 cost per image?

Billing is per token, at $30 per 1M image output tokens as of 10 November 2026. OpenAI's calculator estimates a 1024x1024 image at about $0.025 at low quality, $0.043 at medium, $0.053 at high and $0.211 at max, before input tokens. Wide and tall 1536x1024 images cost slightly less per image at the same quality setting.

Can GPT Image 2.6 make SVG files?

Yes, through the generations endpoint with output_format set to svg and n set to 1. The API makes a reference image first and then writes static SVG code with editable paths and colors, which suits icons and simple logos. SVG output does not stream, cannot come from the edit endpoint, and cannot be used as an input image for edits.

Does GPT Image 2.6 support the Batch API?

No. The model pages for gpt-image-2.6-flare and gpt-image-2.6-sunburst list only the image generation and image edit endpoints, and OpenAI's Batch price table covers gpt-image-2 instead. If a half-price overnight batch matters more than the new features, compare gpt-image-2 Batch pricing against 2.6 standard rates for your quality setting.

Can I still use GPT Image 2.5?

Yes. OpenAI's image generation guide still uses gpt-image-2.5-sunburst and gpt-image-2.5-flare in its main examples and lists the same token rates as 2.6. At low and medium quality, OpenAI's calculator gives 2.5 fewer output tokens per image, so it is cheaper for drafts, while 2.6 adds SVG and GIF output and the complexity control.

Sources

• OpenAI: Introducing Images 2.6
• OpenAI: Image generation guide and cost calculator
• OpenAI: API pricing
• OpenAI: gpt-image-2.6-flare model page
• OpenAI: gpt-image-2.6-sunburst model page