GPT Image 2.5 Flare model guide
A practical guide to GPT Image 2.5 Flare on GPT Image Hub: what changes compared with GPT Image 2 (xhigh and max quality, custom resolutions up to 4K, transparent backgrounds, lower latency), measured output tokens and timings, and when to choose Sunburst instead.
GPT Image 2.5 Flare is OpenAI's default GPT Image 2.5 model for fast, high-quality everyday image generation. OpenAI positions it above GPT Image 2 in quality at roughly half the latency, billed at the same token rates; in GPT Image Hub's own tests a 1536×1024 image at high quality used 1,372 output tokens and 27 seconds, against 5,488 tokens and 78 seconds for GPT Image 2 at the same size. Choose it for most new work; choose Sunburst when editing precision matters most.
Visual examples
Visual examples
A clean commercial scene produced at the default quality.
Everyday hero image
A clean commercial scene produced at the default quality.
1536x1024 product hero on concrete, soft window light, headline space.Transparent sticker sheet
Cutout-ready assets exported as PNG with no background.
Six flat pastel coffee-cup stickers, transparent background.Quality ladder
Draft at auto, approve, then render once at xhigh.
Re-render the approved composition at xhigh quality.Best use cases
- Everyday generation and editing where speed and consistent quality matter more than maximum precision.
- Draft-to-final workflows that move from auto or high to xhigh only for approved compositions.
- Transparent PNG or WebP assets such as cutouts, stickers, icons, and overlays.
- Wide and tall custom resolutions inside the 1:3 to 3:1 range.
Where to be careful
- Edits where precision matters most; OpenAI recommends GPT Image 2.5 Sunburst for those.
- Resolutions above 2560×1440, which OpenAI labels experimental.
Strengths
Adds the xhigh and max quality settings on top of low, medium, high and auto, which earlier GPT Image models stop at high.
Accepts custom WIDTHxHEIGHT sizes: multiples of 16, aspect ratio between 1:3 and 3:1, no edge above 3840 px, and a total of 655,360 to 8,294,400 pixels.
Supports transparent backgrounds when the output format is PNG or WebP, which GPT Image 2 does not.
OpenAI describes it as higher quality than GPT Image 2 at about 50% lower latency, billed at the same per-token rates.
Use cases
Fast everyday generation
Campaign drafts, social creatives, and thumbnails that benefit from lower latency without losing quality.
Transparent assets
Cutouts, icons, and overlays exported as PNG or WebP with a transparent background.
Custom resolutions
Banner and poster formats generated at an exact pixel size instead of a fixed preset.
Workflow
Draft at auto or high
Lock composition, subject, and on-image text at the default quality.
Raise quality for finals
Switch to xhigh or max only for the approved direction; higher settings cost more output tokens.
Set format last
Pick PNG or WebP for transparency, JPEG for the fastest opaque exports.
Prompting guidance
- Keep the prompt structure you already use for GPT Image 2; the 2.5 models accept the same text and reference inputs.
- Reserve xhigh and max for final renders: they produce more image output tokens, which is what OpenAI bills.
- State the output format explicitly when you need transparency: background transparent plus PNG or WebP.
- For custom sizes, choose multiples of 16 and stay within 1:3 to 3:1.
Copy-ready prompts
Create a 1536x1024 product hero of a matte black wireless speaker on a pale concrete plinth, soft north-window light, shallow depth of field, blank space on the right for a headline.
Generate a transparent-background PNG sticker set of six hand-drawn coffee cups, thick outlines, flat pastel fills, no shadows, evenly spaced on one sheet.
Edit the reference image: keep the model's pose and outfit, replace the studio backdrop with a rainy Tokyo street at dusk, match the reflections on the pavement to the new lighting.
Example outputs generated on GPT Image Hub
Official sources
FAQ
Is GPT Image 2.5 Flare available on GPT Image Hub?
Yes. GPT Image 2.5 Flare is selectable in the GPT Image Hub generator with 1K, 2K and 4K tiers and the full aspect-ratio set. Its credit cost per image is listed on the pricing page.
How is Flare different from Sunburst?
OpenAI recommends Flare for fast, high-quality everyday image generation and Sunburst for workflows where editing precision matters most. Both support the same quality settings and sizes; Sunburst takes longer to generate.
Does GPT Image 2.5 cost more than GPT Image 2?
OpenAI bills both GPT Image 2.5 models at the GPT Image 2 token rates, but Flare at high quality produced far fewer output tokens than GPT Image 2 in GPT Image Hub's tests (1,372 versus 5,488 for a 1536×1024 image), so it is cheaper per image here. Higher settings such as xhigh and max, and 4K sizes, use more output tokens; the pricing page lists the Hub credits per tier.
Use this model
Start from a prompt template or compare models first
Use the generator when you already know the model, or compare all supported models when the prompt needs a different balance of speed, detail, and reference inputs.