Launched in November 2026, GPT Image 2.6 generates images 50% faster than before, with Flare finishing a low-quality image in as little as six seconds. It suits teams iterating on posters, product shots and animated mascots, and it succeeds the September 2026 release in the GPT Image line.
What changed
GPT Image 2.6 is OpenAI's image generation model, released November 4, 2026, with Sunburst reading 1259 Elo on the Artificial Analysis Text-to-Image Arena. It ships as Flare and Sunburst API snapshots and adds GIF output, SVG export and a complexity setting.
Where it sits
- $13.50/M$ per 1M tokensBlended price (3:1)Lower is better#86 / 95peer median $1.69/Mvendor price, checked by Hokai
Pricier than 91% of the 95 GA models with a published price, and one of 37 that document a zero-data-retention option.
Ranks are against GA models on Hokai that publish the same figure; ties share a rank.
Provider: OpenAI · Family: GPT Image
Input modalities: text, image · Output: image
About GPT Image 2.6
OpenAI released GPT Image 2.6 on November 4, 2026, branding it Images 2.6 in ChatGPT and Codex. In the API it ships as two snapshots, gpt-image-2.6-flare and gpt-image-2.6-sunburst. It follows ChatGPT Images 2.5 from September 8, 2026 and GPT Image 2 from April 21, 2026, and OpenAI builds it as part of its GPT Image line. OpenAI's launch post does not describe the architecture, parameter count or training cutoff of the 2.6 pair, so Hokai leaves those fields as not disclosed.
The release is mostly about speed and control. OpenAI says Images 2.6 generates images 50% faster than before, and that Flare can finish in as little as six seconds when quality is set to low. Sunburst is the higher-detail snapshot, aimed at consistency across edits at medium or high quality, while Flare also improves layout quality for posters and product imagery. New in the API are image markup (already in ChatGPT), SVG export, animated GIF output, and a complexity parameter that sets how much planning the model does. OpenAI recommends low complexity for most edits and medium or high for scenes with many elements or constraints. Its GIF examples prompt for a six-frame sprite sheet laid out in a 3 by 2 grid that becomes a looping animation.
Independent data comes from the Artificial Analysis Text-to-Image Arena, a blind human-preference vote read on November 7, 2026. Sunburst at max quality sits at 1259 Elo (interval 1244 to 1274, 4,947 votes) and Flare at max quality at 1228 Elo (1216 to 1240, 6,849 votes). The 2.5 snapshots read 1227 for Sunburst and 1208 for Flare, so 2.6 Flare and 2.5 Sunburst overlap within their intervals and should be treated as level. Rivals on the same board include Nano Banana 2.1 at 1176, Grok Imagine Image 2 at 1166 and FLUX 3 at 1117. OpenAI publishes no SWE-bench, GPQA or MMLU scores for an image model, so those fields stay empty.
Billing is per token, with the same rates for Flare and Sunburst and unchanged from 2.5. OpenAI publishes no fixed token count per image, so a flat per-image price does not exist; cost moves with size, quality and complexity. In ChatGPT and Codex the model is available worldwide on desktop, mobile and web across all subscription tiers, and OpenAI names Figma Weave, Picsart, fal, Higgsfield, OpenArt and Sable among early-access users.
OpenAI's system card compares 2.6 with the 2.5 baselines on a policy test set. The share of unsafe generations that were not blocked is 1.72% for Sunburst (1.92% for 2.5 Sunburst) and 2.14% for Flare (1.49% for 2.5 Flare). OpenAI says the 2.6 series performs on par with 2.5 and that the minor numerical regressions are not statistically significant. Weights are closed and the model is API-only, so it cannot be self-hosted or fine-tuned.
Pick Sunburst for iterative edits where a character, product or brand detail must survive several rounds, and Flare for fast drafts and layout-heavy work. Teams that need open weights should look at FLUX.2, and teams chasing a distinctive illustrated look may prefer Midjourney.
Pricing
Token billing, identical for gpt-image-2.6-flare and gpt-image-2.6-sunburst and unchanged from the prior release: text prompt input $5 per 1M tokens ($1.25 cached), reference-image input $8 per 1M ($2 cached), generated image output $30 per 1M. OpenAI publishes no flat per-image price. Batch rates were not read for this release.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.240 | $0.030 | $0.270 |
| Support reply | $0.016 | $0.0090 | $0.025 |
| One coding agent run | $1.60 | $0.600 | $2.20 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- Flare and Sunburst snapshots: Flare is the fast snapshot (as little as six seconds at low quality) and Sunburst the higher-detail one, both billed at the same token rates.
- Animated GIF output: A single prompt or a hand-drawn sketch can become a looping GIF for mascots, reactions and storyboards.
- SVG export: Designs can be exported as SVG so the same output works across websites, apps and print.
- Image markup in the API: Developers can mark the exact region to change, a tool that was already in ChatGPT before the 2.6 release.
- Complexity parameter: A new setting lets developers choose how much planning the model does, from quick low-complexity edits up to crowded multi-element scenes.
Pros
- Sunburst at max quality is the top-rated OpenAI image snapshot on the Artificial Analysis arena.
- GIF output and SVG export cover animated and vector use cases that earlier GPT Image releases did not offer in the API.
- Token rates did not rise with the release, so any speed gain comes at no extra rate.
Cons
- No flat per-image price: cost depends on size, quality and complexity because OpenAI does not publish a token count per image.
- Closed weights and API-only access mean no self-hosting or fine-tuning, unlike FLUX.2.
- Flare's unsafe-not-blocked rate in the system card (2.14%) is above the prior Flare baseline (1.49%), although OpenAI says the difference is not statistically significant.
Benchmarks
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does GPT Image 2.6 cost in 2026?
OpenAI bills per token, and its API pricing page on November 7, 2026 shows the same rates for Flare and Sunburst: $5 per 1M text input tokens ($1.25 cached), $8 per 1M image input tokens ($2 cached) and $30 per 1M image output tokens. These match the 2.5 rates. OpenAI sets no flat per-image price, but Artificial Analysis lists about $210.72 per 1,000 images at 1024x1024 with default settings. Inside ChatGPT and Codex the model is part of the subscription rather than billed per image.
Should you call Flare or Sunburst in the API?
Choose Flare when speed matters: OpenAI puts its low-quality floor at six seconds and says it improves layout for posters and product imagery. Choose Sunburst for detail and consistency across edit rounds at medium or high quality. Both bill the same, so the decision is latency against fidelity, and Sunburst also reads higher on the Artificial Analysis arena at max quality.
What should you use instead of GPT Image 2.6?
Google's Nano Banana 2.1 is the nearest rival on the Artificial Analysis board and is the pick when cost per image weighs more than peak quality. FLUX.2 from Black Forest Labs suits teams that need open weights to self-host or fine-tune. Midjourney remains the choice for a distinctive illustrated look, and Grok Imagine Image 2 fits teams already building on xAI.
What separates GPT Image 2.6 from ChatGPT Images 2.5?
It is the same tier one version up. OpenAI adds GIF output, SVG export, API image markup and the complexity parameter, while token rates stay unchanged. On the Artificial Analysis arena Sunburst at max quality gained 32 Elo over 2.5 Sunburst and Flare gained 20 over 2.5 Flare, with 2.6 Flare and 2.5 Sunburst level within their intervals.
How do you get started with GPT Image 2.6?
In ChatGPT or Codex there is nothing to set up: ask for an image and Images 2.6 handles it on every subscription tier. In the API, create an OpenAI API key and call the image generation or image edit endpoint with the model set to gpt-image-2.6-flare or gpt-image-2.6-sunburst, choosing a quality level and, for edits, a low complexity setting. The first request bills from the first token.
Top Alternatives
- Nano Banana 2.1: Pick Nano Banana 2.1 if cost per image matters most, since Artificial Analysis lists it far cheaper per 1,000 images; pick GPT Image 2.6 for the higher arena Elo and GIF or SVG output.
- FLUX 3: Pick FLUX 3 if your workflow needs video and audio output alongside images; pick GPT Image 2.6 for image, GIF and SVG output with the higher arena Elo.
- FLUX.2: Pick FLUX.2 if you need open weights you can self-host or fine-tune; pick GPT Image 2.6 if you want the higher arena Elo and built-in ChatGPT access.
- Grok Imagine Image 2: Pick Grok Imagine Image 2 if you already build on xAI; pick GPT Image 2.6 if you want markup-guided edits and SVG export through the API.