GPT-Image-2.6: Flare vs Sunburst, GIFs & Pricing

By Versus Tools AI Team · Published · Updated · 9 min read · AI Models & News

We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools

Quick answer. OpenAI Images 2.6 (GPT-Image-2.6) was announced Nov 4, 2026 and is in ChatGPT, Codex, and the API. Flare is the faster model. Sunburst is the higher-quality model. Both use the same API token prices. We did not test it.

OpenAI Images 2.6, also called GPT-Image-2.6, is the image model OpenAI announced on Nov 4, 2026. It is in ChatGPT and Codex across all subscription tiers, on desktop, mobile, and web, and in the API. The announcement said it was available “globally today”. Updated Nov 17, 2026.

This page is based on official documentation and public reports. We did not generate images with GPT-Image-2.6, GPT-Image-2.5, or Seedream 5.5. A per-image dollar figure is not stated. Cost depends on size and quality.

How this page was checked

We read OpenAI’s Nov 4, 2026 announcement, the image generation guide for gpt-image-2.6, and the API pricing page. The predecessor on those pages is GPT-Image-2.5, which also has Flare and Sunburst. The dated rival is Seedream 5.5, from ByteDance’s Seed blog on Nov 9, 2026. Scores in that post are ByteDance’s, and they compare against GPT Image 2.5 Sunburst, not 2.6. Gemini image models (the Nano Banana line) and Midjourney are named only as alternatives. Check each vendor’s docs. No feature claims or numbers for those two are copied here.

Launches are listed on the AI model release timeline and on best AI models. Anthropic’s Nov 10 workplace beta is Claude Meetings. Later comparisons that use the same source rule: Seedance 2.5 Flash vs Gemini Omni 1.2 Flash and Claude Fable 5.2 vs GPT-6.1 Terra.

What is new in OpenAI Images 2.6

The API has two ids. gpt-image-2.6-flare is for fast iteration. gpt-image-2.6-sunburst is for higher-quality generation and greater consistency across edits. OpenAI says images generate 50% faster than before. The announcement does not name the system that 50% is measured against.

OpenAI also says characters, products, and brand details stay consistent across editing rounds. And it says people create 4 billion images per week across ChatGPT, Codex, and the API. That volume figure is OpenAI’s.

GIF output, SVG on generations, markup on edits, and a complexity parameter are the new API controls below. The Standard token price matches GPT-Image-2.5 Flare, GPT-Image-2.5 Sunburst, and GPT-Image-2. The list price is not the change.

Key facts

OpenAI Images 2.6
AnnouncedNov 4, 2026, “globally today”. This page updated Nov 17, 2026.
Also calledGPT-Image-2.6
ChatGPT and CodexAll subscription tiers, desktop, mobile, and web
API idsgpt-image-2.6-flare, gpt-image-2.6-sunburst
Speed50% faster than before, per OpenAI. Flare at low quality: as little as about 6 seconds
EditsCharacters, products, and brand details stay consistent across rounds, per OpenAI
GIFoutput_format: gif on generations or edits. 2 to 16 frames
SVGoutput_format: svg on generations only
Markupmarkup on POST /v1/images/edits. Already in ChatGPT
Complexitylow, medium, high. Default high if omitted. No separate fee
PredecessorGPT-Image-2.5, Flare and Sunburst, same Standard token prices
Volume4 billion images a week, OpenAI’s figure
API noteMay require Organization Verification. safety_identifier recommended. user is deprecated
Tested by usNo

Flare vs Sunburst

The two ids share one Standard price. Choose by the job.

gpt-image-2.6-flaregpt-image-2.6-sunburst
Use caseFast iteration. Also the id OpenAI ties to layout quality for posters, product imagery, and professional visualsHigher-quality generation and greater consistency across edits, for detailed work and editing precision
SpeedAbout 6 seconds is this id with quality set to low. The 50% line is stated for Images 2.6 generallyThe about-6-second line is not stated for Sunburst
Quality setting in the docsLow, for that fast timing. Every Flare call is not required to use lowMedium or high, where OpenAI describes the consistency advantage
Token priceSame Standard card as SunburstSame Standard card as Flare

OpenAI’s GIF example is a Sunburst lantern at frame_count 9 and frame_duration_ms 200, a 1.8 second loop. The docs do not say every GIF must use Sunburst. The sample further down uses Flare, with those same frame settings, because fast iteration is the Flare job.

API pricing

Per 1 million tokens, from OpenAI’s pricing page. Identical for gpt-image-2.6-flare and gpt-image-2.6-sunburst. Batch is half of Standard. The same Standard numbers apply to gpt-image-2.5-flare, gpt-image-2.5-sunburst, and gpt-image-2.

StandardBatch
Image input$8.00$4.00
Image output$30.00$15.00
Cached image input$2.00$1.00
Text input$5.00$2.50
Cached text input$1.25$0.625

gpt-image-1.5 image output is $32 per million tokens. gpt-image-1-mini is $2.50 image input and $8 image output per million tokens. Other cells for those older ids are not copied here. The $30 versus $32 output rate is one published cell, not a full invoice comparison.

Per-image cost depends on size and quality. OpenAI points to the calculator in the image generation guide. This page does not state a per-image dollar figure. complexity has no separate fee. Higher complexity can increase latency.

API use may require API Organization Verification. safety_identifier is recommended. The user parameter is deprecated.

How to make a GIF with GPT-Image-2.6

GIF creation is in the API, and OpenAI’s announcement also makes it available in ChatGPT. The prompt below is an illustration. OpenAI’s published example is a Sunburst lantern at 9 frames and 200 milliseconds, a 1.8 second loop.

Set output_format to gif on POST /v1/images/generations or POST /v1/images/edits.

  • frame_count is 2 to 16. Default 9.
  • frame_duration_ms is 20 to 2000 milliseconds per frame, rounded to 10 milliseconds. Default 200.
  • size describes each frame.
  • Frame count, times dimensions, times n, must fit the pixel budget in the guide. The budget number is not restated here.
  • GIF output does not support streaming, partial images, or custom output_compression.

Still reference images go through the edits endpoint. The sample is a generations call for a text prompt only.

import base64
import json
import urllib.request

body = json.dumps({
    "model": "gpt-image-2.6-flare",
    "prompt": "A paper lantern glowing, then turning slightly",
    "output_format": "gif",
    "frame_count": 9,
    "frame_duration_ms": 200,
}).encode()

req = urllib.request.Request(
    "https://api.openai.com/v1/images/generations",
    data=body,
    headers={
        "Authorization": "Bearer $OPENAI_API_KEY",
        "Content-Type": "application/json",
    },
    method="POST",
)
with urllib.request.urlopen(req) as res:
    payload = json.load(res)

image_bytes = base64.b64decode(payload["data"][0]["b64_json"])
with open("lantern.gif", "wb") as f:
    f.write(image_bytes)

Replace the token before you run it. The snippet uses gpt-image-2.6-flare, output_format, frame_count, frame_duration_ms, and writes b64_json to a .gif. It does not set quality, size, or n. ChatGPT is the non-API path from the announcement. App button labels are not described here.

Coco Mao, CEO and co-founder at OpenArt, in the announcement: “GIFs are the native language of social media, and GPT‑Image‑2.6 lets every OpenArt creator speak it fluently from a single prompt.”

New API controls

Complexity

complexity is low, medium, or high. Omitted, it defaults to high. It sets how much the model plans before rendering, separate from quality, with no separate fee. OpenAI recommends low for most quick edits, and medium or high for dense scenes, text, and layouts. Higher complexity can increase latency. Quality low is the Flare timing knob. Complexity low is the planning knob. The guide does not say they must be paired.

Markup

Markup is already in ChatGPT and is on the API as markup on POST /v1/images/edits. The annotation points at what should change. It applies to the first input image. The limit is 16,777,216 pixels, and the annotation file can be at most 50 MB. Do not combine markup with mask.

SVG output

output_format: svg is generations only, with n of 1, and without streaming, partial images, or custom compression. The file is a static vector: no scripts, no external resources, no embedded rasters. The edits endpoint does not support SVG.

👍 Pros

  • OpenAI says generation is 50% faster than before, and Flare at low quality can return an image in as little as about 6 seconds
  • GIF, SVG, markup, and complexity are documented. Complexity has no separate fee
  • Flare and Sunburst share one Standard token card, matching GPT-Image-2.5

👎 Cons

  • Per-image cost depends on size and quality. This page does not state a per-image dollar figure
  • GIF skips streaming, partial images, and custom output compression. SVG is generations-only
  • Seedream 5.5's comparison targets GPT Image 2.5 Sunburst, not 2.6, and the test is ByteDance's

Images 2.6 versus Seedream 5.5

The head-to-head, including Volcano Ark yuan prices, China access, GIF versus RGBA, and the Artificial Analysis boards, is GPT Image 2.6 vs Seedream 5.5. Seedream 5.5 is the Nov 9, 2026 rival from ByteDance’s Seed blog. Cells below are ByteDance’s claims unless noted. A Pro version will follow. No Pro date is printed, and this page does not give Seedream prices.

ByteDance describes pixel-level preservation across up to 100 editing rounds, native RGBA with transparent backgrounds, more realistic images, and clearer small text in Chinese and English. Its own GSB tests say Seedream 5.5 leads GPT Image 2.5 Sunburst and Nano Banana 2.1 on text structure and text-to-image realism. That target is 2.5, not 2.6. ByteDance’s text figures are 0.4% severe corruption and 1.7% minor. Public reports name a rollout through Dreamina and Jimeng, Doubao, CapCut, BytePlus, and Volcano Ark.

Gemini image models (the Nano Banana line) and Midjourney are alternatives only. No features, prices, or scores for them are stated here. Use each vendor’s docs.

GPT-Image-2.6GPT-Image-2.5Seedream 5.5Gemini image / Midjourney
ReleasedNov 4, 2026Predecessor. Calendar date not in the sources used hereNov 9, 2026See the vendor’s docs
VariantsFlare and SunburstFlare and Sunburst5.5 now. A Pro version will followSee the vendor’s docs
Token priceImage output $30 per 1M tokens. Full card aboveSame Standard card as 2.6Not disclosed hereSee the vendor’s docs
Speed50% faster than before. About 6 seconds is Flare at low qualityBaseline of the 50% is not namedNo seconds figure in the Seed post we usedSee the vendor’s docs
GIF2 to 16 framesNot stated hereNot stated hereSee the vendor’s docs
EditsConsistent characters, products, and brand details, per OpenAINo separate round count hereUp to 100 rounds, pixel-level, per ByteDanceSee the vendor’s docs
TextNo OpenAI corruption rateThe model ByteDance scored is 2.5 SunburstSevere 0.4%, minor 1.7%, ByteDance’s figuresSee the vendor’s docs
Transparencybackground: transparent on PNG, WebP, and SVG, per the image prompting guide read Nov 17, 2026Not disclosed hereNative RGBA, transparent backgroundsSee the vendor’s docs
Offered inChatGPT, Codex, and the APIShared price card. Surfaces not restatedDreamina and Jimeng, Doubao, CapCut, BytePlus, Volcano Ark, per public reportsSee the vendor’s docs

Who should use which

Use Flare for fast iteration, including the poster and product layouts OpenAI ties to that id. The about-6-second timing is quality set to low. Low complexity is the separate recommendation for most quick edits.

Use Sunburst for detailed work and for edit consistency at medium or high quality. OpenAI’s lantern GIF example is Sunburst. Standard token rates match, so price is not the switch.

GPT-Image-2.5 uses the same Standard card. The documented reasons to move are the speed claim, edit consistency, GIF, SVG, markup, and complexity.

Read Seedream 5.5 for ByteDance’s 100-round editing claim, native transparency, or its own text rates. Those tests name GPT Image 2.5 Sunburst and Nano Banana 2.1, not Images 2.6. Public reports put it on Dreamina and Jimeng, Doubao, CapCut, BytePlus, and Volcano Ark. Pro is still to come. No Seedream price is stated here. For Gemini image models and Midjourney, use the vendor’s docs. ChatGPT and Codex cover every subscription tier in the announcement. The API adds the token card, the GIF fields, and a possible organization-verification step.

Latest launches

Claude Meetings is Anthropic’s Nov 10, 2026 Google Meet beta for Team and Enterprise. It is not an image model. Both posts are linked from the AI model release timeline and from best AI models.

Sources

Read on Nov 17, 2026.

Pricing last checked Nov 17, 2026. Per-image cost depends on size and quality. We did not test these models.

Frequently asked questions

Is OpenAI Images 2.6 available on free ChatGPT?

OpenAI's Nov 4, 2026 announcement says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers, on desktop, mobile, and web, and in the API. The announcement does not print a separate exception for the free plan. API use may still require API Organization Verification.

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

gpt-image-2.6-flare is the fast-iteration model. With quality set to low, OpenAI says it can generate an image in as little as about 6 seconds, and it also improves layout quality for posters, product imagery, and professional visuals. gpt-image-2.6-sunburst is for higher-quality generation and greater consistency across edits when quality is medium or high. Standard API token prices are the same for both.

How much does GPT-Image-2.6 cost?

On OpenAI's API pricing page, Standard rates per 1 million tokens are the same for flare and sunburst: image input $8.00, image output $30.00, cached image input $2.00, text input $5.00, and cached text input $1.25. Batch is half of that. The same Standard card applies to gpt-image-2.5-flare, gpt-image-2.5-sunburst, and gpt-image-2. Per-image cost depends on size and quality. OpenAI points to the calculator in the image generation guide. This page does not state a per-image price.

How do I make a GIF with GPT-Image-2.6?

In the API, set output_format to gif on POST /v1/images/generations or POST /v1/images/edits. frame_count runs from 2 to 16 and defaults to 9. frame_duration_ms runs from 20 to 2000 milliseconds per frame, rounded to 10 milliseconds, and defaults to 200. Decode the b64_json field and save a .gif file. OpenAI's announcement also makes GIF creation available in ChatGPT. GIF output does not support streaming, partial images, or custom output_compression.

What is the maximum number of GIF frames in GPT-Image-2.6?

frame_count accepts 2 to 16 frames. The default is 9. OpenAI's Sunburst lantern example uses 9 frames at 200 milliseconds each, which loops in 1.8 seconds. Frame count times the frame dimensions times n must fit the pixel budget in the image generation guide. This page does not restate that budget as a pixel total.

How does Images 2.6 compare with Seedream 5.5?

Images 2.6 launched Nov 4, 2026, with Flare, Sunburst, GIF output, and the token prices above. Seedream 5.5 launched Nov 9, 2026. ByteDance says it preserves pixels across up to 100 editing rounds, outputs native RGBA with transparent backgrounds, and in its own GSB tests leads GPT Image 2.5 Sunburst and Nano Banana 2.1 on text structure and text-to-image realism. That comparison is against 2.5, not 2.6, and the scores are ByteDance's. We did not test either model. Yuan pricing, China access, and the Artificial Analysis boards are on the separate GPT Image 2.6 vs Seedream 5.5 page. Seedream prices are not restated here.