GPT Image 2.5: Flare vs Sunburst, Settings and Credit Costs
Flare or Sunburst? How the two GPT Image 2.5 versions differ, which quality and resolution to pick, what each costs in credits, and when GPT Image 2 is cheaper.
Flare or Sunburst? The Short Answer
Start with Flare. It's the default version of GPT Image 2.5 in the Image Lab, it's the quicker of the two, and it handles everyday images like social posts, thumbnails and drafts. Switch to Sunburst to finish an asset where an edit has to land exactly, like a new background behind a product whose label can't change.
Cost doesn't decide it. Both versions charge the same credits and share every option: 1K to 4K, five quality levels and up to 16 reference images.
What OpenAI Released
OpenAI launched GPT Image 2.5 on September 8, 2026, as ChatGPT Images 2.5 inside ChatGPT and as two API models, gpt-image-2.5-flare and gpt-image-2.5-sunburst. It promises sharper detail, more natural lighting and texture, more recognizable subjects from reference photos, and edits that hold up over several rounds.
OpenAI calls Flare the default for most applications and says it delivers higher quality than GPT Image 2 at 50% lower latency. Sunburst trades speed for tighter control across edits and is aimed at campaign creative and polished product shots (OpenAI's announcement). On Spicy Studio you need no OpenAI account or API key.
What Each Image Costs
Credits per image at 1K · 2K · 4K, identical for Flare and Sunburst:
Low: 4 · 9 · 15
Medium: 9 · 23 · 38
High: 32 · 56 · 90
xhigh: 55 · 95 · 160
Max: 120 · 210 · 350
Every reference image adds 1 credit, so a 2K high edit with three uploads comes to 59. The top levels climb fast: a single 4K max image uses more than half of the 600 monthly credits in Spicy Basic (€10).
New accounts get 25 free credits without a card, enough for six 1K low drafts. Plans and top-ups are on the pricing page.
Quality, Resolution and Aspect Ratios
Quality: use low to test layouts, medium (the default) for most web images and high for text-heavy finals such as posters and packaging. xhigh and max are new in 2.5 (earlier GPT Image models stop at high); save them for dense small print or large prints. Higher levels also take longer to render.
Resolution: 1K for drafts and online use, 2K for hero images, 4K for print. OpenAI's documentation calls output above 2560×1440 experimental, so inspect 4K renders closely. At high quality, a 2K render plus Image Upscale to 4K costs 66 credits, against 90 for native 4K.
Aspect ratios: 15 of them, from 3:1 to 1:3, including 2:1, 1:2 and 9:21. Settle the final ratio first, because a banner and a story need different compositions.
A Flare-to-Sunburst Workflow
Say you're making a launch image for a canned cold brew called Northline.
1. Explore on Flare at 1K medium (9 credits). “A matte navy can of cold brew on wet slate, early side light, condensation on the metal, the can reads ‘NORTHLINE COLD BREW’ in thin white capitals.” Quote every word that should appear.
2. Add references. Upload the real can and the logo (1 credit each) and name them by order: “the can from the first image, the logo from the second.”
3. Polish on Sunburst at 2K high (58 credits with the two references). Switch to Sunburst and state what stays fixed: “Keep the can, label and lighting exactly as they are. Replace the background with a sunlit kitchen counter.” OpenAI says 2.5 is better at changing only what you ask for.
4. Make one change per edit, then proofread. OpenAI's documentation still warns that text placement can slip. More techniques are in the prompt engineering guide.
GPT Image 2.5 or GPT Image 2?
For new work, start with 2.5. It adds 4K, xhigh and max, and five extra aspect ratios, and it costs a little less at matching settings: 9 credits instead of 10 at 1K medium, 56 instead of 60 at 2K high.
GPT Image 2 doesn't charge for references, though. Both accept 16, but with ten references at 2K medium, GPT Image 2 costs 25 credits and 2.5 costs 33. The GPT Image 2 guide has the details. Looking for the older GPT-4o image generator? Here's what replaced it.
When to Use Something Else
Mature themes: GPT Image models filter prompts and outputs under OpenAI's content policy. For 18+ work with fictional or consenting adults, see Spicy UNCENSORED.
Many references at a flat price: Nano Banana 2.1 takes up to 14 for 12 credits at 2K.
Motion: animate the result from a first frame in Seedance 2.5, or see the video model comparison.
Frequently Asked Questions
What's the difference between Flare and Sunburst in GPT Image 2.5?
+
Flare is faster and is OpenAI's default for most images. Sunburst is slower but built for precise edits on finished work. On Spicy Studio they cost the same credits and have identical settings.
When did OpenAI launch GPT Image 2.5?
+
September 8, 2026. It arrived as ChatGPT Images 2.5 in ChatGPT and as the gpt-image-2.5-flare and gpt-image-2.5-sunburst models in OpenAI's API.
What does GPT Image 2.5 cost?
+
On Spicy Studio, from 4 credits (1K, low) to 350 credits (4K, max), plus 1 credit per reference image. In OpenAI's API, as of October 2026, both versions cost $8 per million image input tokens and $30 per million image output tokens, the same rates as GPT Image 2.
Does GPT Image 2.5 support 4K?
+
Yes, at 15 to 350 credits depending on quality. OpenAI calls output above 2560×1440 experimental, so check 4K results closely.
Is GPT Image 2.5 free to try?
+
New Spicy Studio accounts get 25 free credits with no card and no OpenAI key. ChatGPT's free plan also includes Images 2.5, with separate usage limits for image creation.
Want to run this yourself? Every new account starts with free credits.
Try GPT Image 2.5