Tooba Siddiqui
October 2, 2026 • Updated October 2, 2026
11 mins Read
GPT Image 2.5 is OpenAI's latest image generation and editing model that comes in two versions, Flare for speed and Sunburst for precision, and it's available in ChatGPT on every plan, through the OpenAI API and on ImagineArt. Compared with GPT Image 2, it's faster, edits more precisely and renders short text more accurately.
This guide covers what's new, the full specs, pricing, how to use it for free, and how to get the best results.
What Is GPT Image 2.5?
GPT Image 2.5 is OpenAI's image generation and editing model. It creates images from text prompts, edits images you upload, and works from up to 16 reference photos.
You can use OpenAI GPT Image 2.5 in three main ways:
- In ChatGPT: on every plan, including free, on web, desktop and mobile
- On ImagineArt: alongside 30+ other image models in the AI image generator
Sunbrust and Flare share the same features and pricing. The difference is how long each spends on an image and how closely it follows detailed instructions.
Key Features of GPT Image 2.5
GPT Image 2.5 is a refinement of GPT Image 2. The biggest changes are the two-model split, more precise editing, and higher quality ceilings.
Two Models: Flare and Sunburst
Flare is the default. It produces better images than GPT Image 2 at up to 50% lower latency. Sunburst spends more time on each image for tighter edits and better detail. In our tests, Sunburst won five of seven jobs, but Flare was the better pick for product photos and realistic street scenes. See the full Sunburst vs. Flare test results.
Targeted Editing and Reference Images
GPT Image 2.5 edits the area you describe, while keeping the rest of image same. Reference photos also hold up better. When you upload a photo of a person or product, GPT Image 2.5 keeps them recognizable in new settings and styles, carrying lighting, texture and distinctive features through more consistently than GPT Image 2. You can add up to 16 reference images.
Better On-Image Text
Small text rendering is one of the clearest improvements. In our tests, both models rendered every word of a four-line event poster correctly, including something as minute as the separators. However, smaller print, like a product label, was had certain spelling and legibility issues when created with Flare, while GPT image 2.5 Sunburst got every word right.
New Quality Tiers
GPT Image 2.5 adds two quality tiers above GPT Image 2's previous ceiling of high. The full range is now auto, low, medium, high, xhigh, and max. Higher tiers produce more detail but take longer and cost more, so the tier you choose has a big effect on price. On ImagineArt, a 1K image costs 3 credits at low quality and 120 credits at max.
New ChatGPT Tools: Sketch, Templates and Image Comments
ChatGPT added four tools alongside GPT Image 2.5:
- Sketch: draw a rough layout, and the model follows it
- Templates: start from a ready-made layout for common formats
- Image comments: click on part of an image and describe the change, instead of rewriting the whole prompt
- Shareable prompts: share the prompt behind an image so others can reuse it
GPT Image 2.5 Specs
GPT-Image-2.5 outputs up to 3840 x 2160 pixels, in PNG, JPEG or WebP, with six quality tiers. Here are the full specs:
| Spec | Detail |
|---|---|
| Developer | OpenAI |
| Release date | September 8, 2026 |
gpt-image-2.5-flare, gpt-image-2.5-sunburst | |
| Max resolution | 3840 x 2160 (3840px max on the long edge) |
| Standard sizes | 1024 x 1024, 1536 x 1024, 1024 x 1536, plus custom sizes in multiples of 16px |
| Aspect ratios | 1:3 to 3:1 |
| Quality tiers | auto, low, medium, high, xhigh, max |
| File formats | PNG, JPEG, WebP |
| Transparent backgrounds | Supported |
| Reference images | Up to 16 per edit |
| Provenance | C2PA metadata and an invisible SynthID watermark |
Sizes above 2560 x 1440 are still labeled experimental, and dense requests at high resolution can take up to two minutes.
Output Resolution, File Formats and Downloads
GPT Image 2.5 supports resolutions up to 3840 x 2160, but the size you get depends on where you use it.
- ChatGPT: by default ChatGPT generates 1024×1024 pixel image in square format, unless mentioned otherwise in the image prompt.
- ImagineArt: choose 1K, 2K or 4K resolution and aspect ratio before generating. Higher resolutions cost more credits.
File formats: PNG, JPEG and WebP. Use PNG for transparent backgrounds or anything you'll edit further, and JPEG or WebP for smaller files on the web. Transparent backgrounds work in PNG and WebP.
If you need to improve the resolution of the AI-generated image, use the ImagineArt AI image upscaler.
GPT Image 2.5 vs. GPT Image 2: What Changed
GPT Image 2.5 is faster, edits more precisely, and reaches higher quality than GPT Image 2.
| GPT Image 2 | GPT Image 2.5 | |
|---|---|---|
| Models | One | Two (Flare and Sunburst) |
| Speed | Baseline | Up to 50% lower latency (Flare) |
| Top quality tier | high | max (plus xhigh) |
| Editing | Often regenerated more of the frame | Changes only the area you describe |
| Reference images | up to 16 | up to 16 |
| Short text | Good | More accurate |
GPT Image 2.5 Pricing
GPT Image 2.5 pricing depends on where you use it: token-based in the API, credit-based on ImagineArt, and included in every ChatGPT plan. Flare and Sunburst cost the same everywhere.
OpenAI API Pricing
The GPT Image 2.5 API charges per token, at the same rates for both models:
| Token type | Price per 1M tokens |
|---|---|
| Text input | $5.00 |
| Cached text input | $1.25 |
| Image input | $8.00 |
| Cached image input | $2.00 |
| Image output | $30.00 |
GPT Image 2.5 Pricing on ImagineArt
On ImagineArt, GPT Image 2.5 costs between 6 and 120 credits per image at 1K resolution, depending on the quality tier. Flare and Sunburst consume the same amount of credits.
Credits per image:
| Quality | 1K | 2K | 4K |
|---|---|---|---|
| Low | 6 | 7.5 | 12 |
| Medium | 12 | 15 | 24 |
| High | 30 | 37.5 | 60 |
| Xhigh | 51 | 63.75 | 102 |
| Max | 120 | 150 | 240 |
What that costs per image in dollars, at 1K:
| Quality | Ultimate ($55 for 16,000 credits) | Basic ($13 for 3,000 credits) |
|---|---|---|
| Low | ~$0.02 | ~$0.03 |
| Medium | ~$0.04 | ~$0.05 |
| High | ~$0.10 | ~$0.13 |
| Xhigh | ~$0.18 | ~$0.22 |
| Max | ~$0.41 | ~$0.52 |
ChatGPT Plans
GPT Image 2.5 is included in every ChatGPT plan, from free to Pro, with higher usage limits on paid plans. You can create 2 to 3 images per day using free plan.
How to Access GPT Image 2.5
You can access GPT Image 2.5 on four kinds of platforms:
- ChatGPT: on every plan, on web, desktop and mobile, plus ChatGPT Work and Codex
- ImagineArt: in the AI image generator, where you can compare it with other image models and upscale and edit images in one place.
How to Use GPT Image 2.5
How you use GPT Image 2.5 depends on the platform. The steps below cover ImagineArt, ChatGPT and the API.
On ImagineArt
- Open the AI image generator on ImagineArt.
- Select GPT Image 2.5 from the model list, then choose Flare or Sunburst
- Choose a resolution (1K, 2K or 4K) and a quality tier.
- Type your prompt, or upload a reference image to edit.
- Select generate, then download the image, refine it with ImagineArt AI image editor or animate images with AI video generator.
In ChatGPT
- Open ChatGPT and start a new chat.
- Describe the image you want, or upload a photo and describe the change.
- To control the layout, use Sketch to draw a rough composition, or start from a Template.
- To fix one part of an image, comment directly on that area instead of rewriting the prompt.
- Download the image, or share the prompt with a teammate.
GPT Image 2.5 Prompt Tips That Work
The best GPT Image 2.5 prompts name the subject, the placement, any exact text, the lighting and the camera, in that order. This template worked well across our tests:
[Type of image] of [subject], [key action or pose].
[Placement: what goes where, as it appears in the image].
[Exact text in quotes, with its size and position].
[Lighting and its direction]. [Camera, lens and aperture].
[Style]. No logos.
For edits: Change only [one specific thing]. Keep [the subject, layout and lighting] exactly the same.
Tips from our tests:
- Describe left and right as the viewer sees them. Both models read "her left" as the left of the image. Write "on the right side of the image" instead.
- Add camera and lens details for realism. Adding "Shot on a Fujifilm X100V, 23mm, f/2, natural color, no retouching" turned a postcard-style image into a convincing photo. It made a bigger difference than switching models.
- Change one thing per edit. Name it specifically, and say what must stay the same.
- Add "unbranded, no logos" for commercial work. Flare added an Apple logo to a laptop we never asked for.
- Put exact text in quotes and give each line a size: large, medium or small.
- Don't name a famous city unless you want landmarks. A prompt set in Lisbon came back full of trams and tourist signs. Add "an ordinary side street, no landmarks" for a candid look.
For the reasons these work, see our guide to what makes an AI image prompt work.
Limitations
GPT Image 2.5 still has a few weak spots, on both models:
- Left and right: directions are usually read from the viewer's side, even when you mean the subject's
- Edits on uploaded photos: precise on generated images, but in our test, both models reframed an uploaded photo instead of changing only one object
- Unrequested additions: brand logos, landmarks and extra props can appear without being asked for
- Filler text: background text you don't specify comes out as unreadable scribbles
- Synthetic textures: skin and some surfaces can look overly sharpened or processed
- Long text: short lines are reliable, but long passages and very small print still break
Frequently Asked Questions
When was GPT Image 2.5 released?
OpenAI released GPT Image 2.5 on September 8, 2026. It launched in ChatGPT, ChatGPT Work and Codex, and as two API models, Flare and Sunburst.
Is GPT Image 2.5 free?
Yes. GPT Image 2.5 is included in ChatGPT's free plan with a daily limit of 2-3 images per day. ImagineArt offers 100 free daily credits.
Can I use GPT Image 2.5 in ChatGPT?
Yes. GPT Image 2.5 is the image model in ChatGPT on every plan, on web, desktop and mobile. You can generate images, edit uploads, sketch a layout or comment on part of an image to change it.
What resolution does GPT Image 2.5 support?
GPT Image 2.5 supports up to 3840 x 2160 pixels through the API, with aspect ratios from 1:3 to 3:1. ChatGPT's default output is smaller, and ImagineArt offers 1K, 2K and 4K options.
Can I use GPT Image 2.5 images commercially?
Yes, with some caveats. Every image carries C2PA metadata and a SynthID watermark that identify it as AI-generated, and trademark and likeness rules still apply. In the US, purely AI-generated images can't be copyrighted without meaningful human authorship. For more, see our guide to AI image generators for commercial use.
Is GPT Image 2.5 better than GPT Image 2?
Yes, in most ways. GPT Image 2.5 is up to 50% faster, edits more precisely, adds two higher quality tiers and renders short text more accurately. Long text and synthetic-looking textures are still weak spots.
Can GPT Image 2.5 upscale an image?
GPT Image 2.5 can recreate an image at a higher resolution, but it redraws the image rather than enlarging it, so details and text may change. To enlarge an image without changing the design, use a dedicated AI image upscaling tool.
Tooba Siddiqui
Tooba Siddiqui is a content marketer with a strong focus on AI trends and product innovation. She explores generative AI with a keen eye. At ImagineArt, she develops marketing content that translates cutting-edge innovation into engaging, search-driven narratives for the right audience.