
Key takeaways
OpenAI's newest image models are GPT Image 2.6 Flare and Sunburst, released on November 4, 2026. Flare is the quick variant and Sunburst is tuned for detail and consistent edits.
Melius tops this list because GPT Image sits beside Nano Banana, Seedream, Ideogram and many other models on one canvas, so you can switch models on any step without leaving the project.
ChatGPT includes the models on every plan, the free one too, though OpenAI doesn't publish the free limit.
OpenAI bills its API by tokens and publishes no per-image price for 2.5 or 2.6, so the final cost depends on quality and size.
Several platforms still list older versions, so check which one you're getting before you pay.
How we ranked
GPT Image is a family of OpenAI models, and most platforms sell access to it as one option among many. We ranked platforms on which versions they offer today, how well they expose the model's features (reference images, quality, size, transparency), what else you can do alongside it, how clearly you can predict the cost and who owns the output. Everything was checked on each company's own pages on November 12, 2026.
Platforms for GPT Image at a glance
Platform | GPT Image versions | Best at | Starts at |
|---|---|---|---|
Melius | 2.5 Sunburst and Flare, GPT Image 2, Better Fonts on 2.6 | One canvas with many models | $17/mo billed yearly (Creator) |
ChatGPT | 2.6 | Editing by conversation | Free, with Go at $8/mo |
OpenAI API | 2.6 Flare and Sunburst, 2.5, 2 | Building into your own app | Pay by token |
Adobe Firefly | 2.6 Sunburst and Flare | Working next to Photoshop | $9.99/mo (Standard) |
Artlist | 2.6 Flare, Sunburst, GIF and Vector, 2.5, 2 | Creators who also need licensed media | $11.99/mo billed yearly (AI Starter) |
Krea | 2.5 and 2.6 | Trying GPT Image beside Krea's own tools | $5/mo billed yearly (Basic) |
Higgsfield | 2.6 Flare and Sunburst | GIF and SVG output | $19/mo billed yearly (Starter) |
fal.ai | 2.6 Flare | Developers who want pay-as-you-go | Pay by token |
1. Melius: Best overall platform for GPT Image
Melius calls itself the AI-native operating system for creatives. Its AI agents for creative work build each image job as connected nodes on a canvas.
What works: GPT Image 2.5 Sunburst at High quality is the default model on an image node, and GPT Image 2.5 Flare and the GPT Image 2 quality tiers are one click away. Each model takes up to 16 connected reference images, and you choose Auto, Low, Medium, High, Extra high or Max quality at 1K, 2K or 4K. A transparent background option makes stickers, logos and cutouts without changing the credit price. Better Fonts, available on Professional and above, uses GPT Image 2.6 Sunburst at High quality to set text in a font you choose. Because the same canvas holds Nano Banana, Seedream, Ideogram and Flux, you can compare GPT Image with another model by switching the model on a node and rerunning just that step. The Melius agent builds the nodes from a plain-language brief, and your images stay private to your workspace. Creator costs $17/mo billed yearly and includes about 300 images a month.
What doesn't: There are usually a few days between a new GPT Image release and the model going live on Melius.
Best for: Creators and teams who want GPT Image alongside other models in one project.
2. ChatGPT: Best for editing by conversation
ChatGPT is where most people first use GPT Image, and Images 2.6 reached every plan on November 4, 2026.
What works: You describe an image, then refine it in the same chat, or mark up an area and describe the change. The models are included on the free plan, and Go is $8 a month, Plus is $20 and Pro starts at $100 with the highest limits. OpenAI's terms say you own the output and assign you its rights, with a note that outputs may not be unique.
What doesn't: OpenAI doesn't publish how many images the free plan allows. There's no canvas, no brand kit and no way to compare other models.
Best for: Anyone who wants GPT Image with the least setup.
3. OpenAI API: Best for building into your own app
OpenAI sells GPT Image 2.6 Flare and Sunburst directly through its API, along with 2.5 and GPT Image 2.
What works: Flare generates quickly, and OpenAI says a low-quality image can take as little as six seconds, while Sunburst is the version for detail and consistent edits. The 2.6 models can output SVG and animated GIFs, accept markup-guided edits and take a complexity setting that controls how much planning happens before drawing. Images can be transparent and run up to 4K. Rates are $8 per million image input tokens and $30 per million image output tokens.
What doesn't: OpenAI publishes token rates and no per-image price for 2.5 or 2.6. The last per-image table it printed covers GPT Image 2 at 1024 by 1024, at $0.006 for low quality, $0.053 for medium and $0.211 for high. Some accounts need organization verification before they can use the models.
Best for: Developers adding image generation to their own product.
4. Adobe Firefly: Best for working next to Photoshop
Firefly mixes Adobe's own models with partner models, and its November 2026 announcement lists GPT Image 2.6 Sunburst and Flare.
What works: Results land in the same app as Adobe's other generative tools, and Standard costs $9.99 a month with 2,000 credits, which Adobe says is up to 400 images.
What doesn't: Adobe's own GPT Image page still says GPT Image 2, so the listed version is inconsistent. Images from partner models spend generative credits and fall outside the commercial-safety promise Adobe makes for its own models.
Best for: Photoshop users who want GPT Image inside their usual workflow.
5. Artlist: Best for creators who also need licensed media
Artlist sells stock footage and music, and its AI platform offers GPT Image 2, 2.5 and 2.6.
What works: The 2.6 models include GIF and Vector outputs. Each edit takes up to 16 references, and sizes run from 1K to 4K. A High-quality 1K text-to-image costs 100 credits. AI Starter is $19.99 monthly or $11.99 billed annually with 16,500 credits, and the license covers commercial use.
What doesn't: GPT Image 2.5 and 2.6 aren't included in any Unlimited plan. High-quality edits cost 100 to 200 credits, so Starter works out to roughly 80 to 160 GPT Image generations a month.
Best for: Video and content creators who already license music and footage.
6. Krea: Best for trying GPT Image beside other tools
Krea lists GPT Image 2.5 and 2.6 Flare and Sunburst on its pricing page.
What works: There's a free allowance of 100 compute units a day. Basic is $9 monthly or $5 billed annually with 5,000 units, and paid plans include a commercial license. Krea's live canvas, style training and other models are in the same account.
What doesn't: GPT Image is one of Krea's costlier models per generation, so the cheaper plans go only so far.
Best for: Designers who already use Krea and want GPT Image as another option.
7. Higgsfield: Best for GIF and SVG output
Higgsfield added GPT Image 2.6 Flare and Sunburst on launch day, November 4.
What works: Outputs include PNG, animated GIF and SVG in preview, with transparent backgrounds, up to 16 references and markup edits. On its Plus plan, a High-quality 2K image costs 2.75 credits, which Higgsfield puts at about $0.13. Starter is $19 a month billed yearly.
What doesn't: Pricing is on a sale that Higgsfield says will change, and its Unlimited window leaves out High quality, 4K, GIF and SVG.
Best for: Creators making animated or vector GPT Image output.
8. fal.ai: Best for pay-as-you-go API access
fal.ai hosts models for developers, and it has a page for GPT Image 2.6 Flare.
What works: There's no subscription. You pay OpenAI's token rates, $8 per million image input tokens and $30 per million output tokens, and quality defaults to high.
What doesn't: It's API only, with no editing interface, and there's no flat per-image price.
Best for: Developers who prefer one provider for many models.
Using GPT Image in Melius
Open a canvas, add an image node and choose a GPT Image model. Pick the quality and size, attach as many as 16 reference images and write the prompt, for example "A flat-lay of a ceramic mug and an open notebook on a wooden desk, soft window light, with the words 'Monday Reset' on the mug." You can also describe the job to the Melius agent and let it build the nodes. When the mug text comes out wrong, change the prompt on that node or try a different model, then rerun that one step. For stickers and logos, turn on the transparent background option, and for text in a specific font, use Better Fonts.
How to choose a platform for GPT Image
Decide whether you want GPT Image alone or next to other models.
Check which version the platform offers, since 2.6 is new and some still list 2.5.
Compare the cost of a high-quality image on each plan, because credits and tokens vary widely.
Look for the controls you need, such as references, transparency or SVG and GIF output.
Read the terms for who owns the results.
Frequently asked questions
What is the best platform to use GPT Image?
Melius is the best platform to use GPT Image in 2026, because GPT Image sits next to many other models on one canvas, with up to 16 references and adjustable quality. ChatGPT is the simplest option, and OpenAI's API suits developers.
Is GPT Image free?
Partly. ChatGPT's free plan includes image creation with limits OpenAI doesn't publish, and Krea gives 100 free compute units a day. API access and most other platforms charge.
What is the difference between GPT Image Flare and Sunburst?
Flare is the faster variant, and OpenAI says a low-quality image can take as little as six seconds. Sunburst is tuned for more detail and more consistent edits at medium and high quality.
Can GPT Image make transparent backgrounds?
Yes. OpenAI's API supports transparent backgrounds in PNG and WebP, and Melius has a transparent background option for stickers, logos and cutouts.
Who owns images made with GPT Image?
OpenAI's terms say you own the output and that it assigns you its rights. On Melius, the company says: "You own the content you create on Melius and can use, publish, and distribute it commercially, subject to the underlying model providers' terms. Melius doesn't claim ownership of your work and doesn't use it to train AI models."