What Is GPT Image 2.6? Flare vs Sunburst, GIF, SVG

GPT Image 2.6 is OpenAI's newest image generation and editing model, released on November 4, 2026. In ChatGPT it is called Images 2.6; in the API it ships as two models, GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Beyond still images, it can return animated GIFs and SVG vectors, follow a markup overlay when editing, and let you set how much the model plans before it draws. You can call both models today through the GPT Image 2.6 API on APIMart.

This guide covers what the release actually changes, how to choose between Flare and Sunburst, where 2.6 costs more than 2.5, and request examples you can copy for each new feature.

Key Takeaways

• GPT Image 2.6 is OpenAI's image model released on November 4, 2026, available as gpt-image-2.6-flare and gpt-image-2.6-sunburst; OpenAI says it generates images 50% faster than the previous generation.
• Flare is the fast default for drafts and volume; Sunburst gives higher quality at medium or high quality and keeps details more consistent across edit rounds. Both take the same parameters and cost the same.
• New in 2.6: GIF animation (2–16 frames on APIMart), SVG export, markup-guided editing in the API and a complexity setting that is separate from quality.
• Token rates match GPT Image 2.5, but low quality now uses about 4.2× and medium about 3.3× the output tokens; high, xhigh and max are unchanged.
• On APIMart a 1024×1024 image costs about $0.020 at low, $0.035 at medium and $0.042 at high quality (Gold price, November 7, 2026).

What Is GPT Image 2.6?

OpenAI announced Images 2.6 on November 4, 2026 and launched it in ChatGPT, Codex and the API on the same day ^[1]. The ChatGPT feature and the API models are the same release under different names: ChatGPT users see "Images 2.6", while developers call one of two model IDs.

Two API Models, One Release

An APIMart API key calls the two API models from your own code. It is billed per use and is separate from any ChatGPT subscription.

Where It Sits in the GPT Image Line

GPT Image 2 arrived in April 2026, GPT Image 2.5 introduced the Flare / Sunburst split in September, and 2.6 keeps that split while adding new output formats. OpenAI is also retiring older models: gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest leave the API on December 1, 2026, and OpenAI names 2.6 Flare or Sunburst as the replacement ^[2].

GPT Image 2.6 at a Glance

The table below reflects the APIMart implementation as documented on November 7, 2026 ^[3].

Flare vs Sunburst: Which One Should You Use?

The two models accept exactly the same request body and cost the same per token, so switching is a one-word change in model. The difference is how they trade speed for detail.

When to Use Flare

GPT-Image-2.6 Flare is built for fast iteration. OpenAI says Flare can return an image in as little as about six seconds at low quality, and that it improves layout for professional visuals such as posters and product images ^[1]. Use it for concept rounds, batch variations, social formats and any job where you will review many candidates and keep a few.

When to Use Sunburst

GPT-Image-2.6 Sunburst is the model to pick when the image has to hold up under scrutiny. OpenAI says that at medium or high quality, Sunburst produces higher image quality and keeps details more consistent from one edit to the next ^[1]. That matters for brand campaigns, product shots that go through several revision rounds and characters that must look the same across a set. Expect it to take longer than Flare for the same request.

A Simple Rule of Thumb

A common pattern is to explore with Flare, then re-run the winning prompt on Sunburst before delivery.

GPT Image 2.6 vs GPT Image 2.5

What Is New in 2.6

How Token Usage Changed

The per-token prices are identical: $30 per million image output tokens and $8 per million image input tokens at list price. What changed is how many tokens each quality level uses ^[4]:

In practice, a 1024×1024 low-quality image costs about $0.005 on 2.5 and about $0.020 on 2.6 at the Gold price. Low quality on 2.6 is no longer the very cheap drafting tier it was on 2.5.

When 2.5 Still Makes Sense

Stay on GPT Image 2.5 if your workload is large volumes of low or medium still images and you do not need GIF, SVG, markup or complexity. At high quality and above, the two versions cost the same, so 2.6 is the better default there. Run the same prompts through both before switching a production pipeline.

New in GPT Image 2.6: GIF, SVG, Markup and Complexity

All four features use the same POST /v1/images/generations endpoint. The examples below are request bodies; add your API key in the Authorization: Bearer header.

Animated GIFs

Set output_format to gif. The model draws all frames on a single sheet and APIMart turns it into an animation, which is why size here means the size of each frame, not the whole image. Omit it or use auto for 512×512 frames; aspect-ratio names such as 1:1 are not accepted.

{
  "model": "gpt-image-2.6-flare",
  "prompt": "a paper lantern swaying gently in the night breeze, seamless loop",
  "output_format": "gif",
  "frame_count": 9,
  "frame_duration_ms": 200,
  "quality": "low"
}

• frame_count runs from 2 to 16 (default 9); frame_duration_ms defaults to 200 and must be at least 20.
• The sheet is laid out as 4×2 for 2–8 frames, 3×3 for 9, 4×3 for 10–12 and 4×4 for 13–16, and must stay within 3840 px per edge. 512×512 frames work at every count, 768×768 at up to 12, and 1024×1024 at none.
• You can animate from text or from up to 16 reference images, but GIFs do not accept masks, markup or output_compression.

At the default 200 ms per frame, 9 frames play for 1.8 seconds and 16 frames for 3.2 seconds per loop.

SVG Vectors

Set output_format to svg and name a text model in svg.model. The image model first draws a reference image, then that text model writes it as SVG code.

{
  "model": "gpt-image-2.6-flare",
  "prompt": "a minimalist bakery logo with a wheat stalk, flat colors",
  "output_format": "svg",
  "svg": { "model": "gpt-5.4-mini" },
  "size": "1:1",
  "quality": "low"
}

SVG is text-to-image only, with n set to 1, and the result is a static vector with a transparent background and no embedded bitmaps. You pay for the reference image plus the text model's actual usage, reported separately as svg_usage. A cheaper writing model such as gpt-5.4-mini is usually enough for icons and logos.

Markup vs Mask Editing

Both tools limit where an edit happens, but they work differently:

{
  "model": "gpt-image-2.6-flare",
  "prompt": "Recolor the striped vase inside the red circle to solid cobalt blue ceramic. Keep everything else unchanged. Remove the red circle.",
  "image_urls": ["https://example.com/kitchen.webp"],
  "markup_url": "https://example.com/vase-circle-overlay.png",
  "quality": "medium"
}

The overlay must match the size of the first image in image_urls, and you cannot send mask_url and markup_url together.

Complexity vs Quality

quality controls how finely the image is rendered; complexity (low, medium or high, default high) controls how much the model plans before rendering. OpenAI recommends low for quick edits such as swapping a background, and medium or high for scenes with many elements, dense text or complex layouts. On APIMart, complexity adds no separate charge. Only the 2.6 models accept it; sending it to 2.5 returns a 400 error.

Cost Examples

Prices below are APIMart Gold prices checked on November 7, 2026 (list price minus 20%). Final charges follow actual token usage; check the model page for current rates.

Still Images

Reference images add input cost: a 1024×1024 input image is 1,024 image input tokens, about $0.0066 at the Gold price.

A Default GIF

A 9-frame GIF with 512×512 frames is drawn on a 1536×1536 sheet and costs about $0.028 at low, $0.050 at medium and $0.060 at high quality. Note that 10–12 frames use a 2048×1536 sheet that costs slightly less than the 9-frame 3×3 sheet, so frame count does not map linearly to price.

A Fixed-Price Option

The gpt-image-2.6-ext route charges a flat price per image for either version: $0.0085 at 1K, $0.03 at 2K and $0.04 at 4K (Gold price, November 8, 2026), with n fixed at 1. See the Ext documentation for its parameters.

Prompt Tips for GPT Image 2.6

Poster and Headline Text

Put the exact wording in quotes, say where it goes and leave room for it: "Headline 'SUMMER SALE' centered in the top third, bold sans-serif, white on a coral background, generous margins." Proofread every generated word before publishing.

Product Shots That Keep the Details

Pass the product photo in image_urls and list what must not change (logo, label text, colors, proportions) before you describe the new scene. Use Sunburst at medium or high quality when the product will be edited several times.

Multi-Round Edits

In each round, describe only the change: "Change the jacket to dark green. Keep the face, pose, lighting and background unchanged." Restating the whole scene invites the model to redraw parts you already approved. Pair this with complexity: "low" for small fixes.

How to Access GPT Image 2.6

1. Get an API key. Sign up on APIMart, add balance and create a key. The same key calls Flare, Sunburst and every other model on the platform.
2. Submit a task. Send a request to POST https://api.apimart.ai/v1/images/generations and read data[0].task_id from the response.
3. Query the result. Poll GET https://api.apimart.ai/v1/tasks/{task_id} every few seconds until the status is completed, then read the file URL from data.result.images[0].url[0].

curl https://api.apimart.ai/v1/images/generations \
  -H "Authorization: Bearer $APIMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.6-sunburst",
    "prompt": "a ceramic coffee cup on a marble counter, soft morning light",
    "size": "4:5",
    "resolution": "2k",
    "quality": "medium"
  }'

Download the files you want to keep within 72 hours. The GPT-Image-2.6 documentation lists every parameter and error message. If you are weighing OpenAI against Google's image models, our GPT Image 2 vs Nano Banana 2 comparison covers the trade-offs that still apply.

FAQs

How long can a GPT Image 2.6 GIF be?

On APIMart a GIF can have 2 to 16 frames. With the default 200 ms per frame, that is 0.4 to 3.2 seconds per loop; you can lengthen playback by raising frame_duration_ms, but the frame count stays capped at 16. For longer clips, use a video model instead.

Can a GPT Image 2.6 SVG include photos or embedded images?

No. The SVG output is a static vector without embedded bitmaps, scripts or external resources, and SVG requests cannot take reference images. It suits logos, icons and flat illustrations rather than photographic scenes.

Can I run GPT Image 2.6 locally?

No. OpenAI has not released the model weights. GPT Image 2.6 is available only through ChatGPT, Codex and the API, including providers such as APIMart.

How do I migrate from GPT Image 1.5 to 2.6?

Change model to gpt-image-2.6-flare or gpt-image-2.6-sunburst and keep your prompts. Review size and quality settings, because 2.6 uses aspect ratios with resolution tiers and its quality levels use different token amounts. OpenAI removes gpt-image-1.5 from its API on December 1, 2026, so plan the switch before then.

Does complexity change the price?

No. On APIMart, complexity does not add a separate charge or extra image tokens. Higher settings can take longer and may introduce changes you did not ask for, so compare low and high on your own prompts.

Can I use markup and a mask in the same request?

No. Choose one: mask_url for an exact region, or markup_url for a visual hint. Both require at least one image in image_urls, and neither works with GIF or SVG output.

What to Watch Next

The next date to watch is December 1, 2026, when OpenAI removes gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest from its API. We will update this article if limits or prices change. In the meantime, start with Flare for drafts, move finals to Sunburst and check the model page before you scale a batch.