# GPT Image 2.5 explained: what's new, Flare vs Sunburst, and how to try it

OpenAI released GPT Image 2.5 on 8 September 2026 as two models: Flare, the fast everyday default, and Sunburst, for work where precision matters most. Compared with GPT Image 2 it adds xhigh and max quality, custom sizes up to 3840 pixels on an edge and transparent backgrounds, at the same token price and about half the latency. This guide covers what changed, how the two models differ, what an image costs, and real outputs from GPT Image Hub.

URL: https://www.gptimagehub.com/guides/gpt-image-2-5
Published: 2026-11-06
Updated: 2026-11-06

## GPT Image 2.5 in one paragraph

GPT Image 2.5 is OpenAI's image model from September 2026, offered as gpt-image-2.5-flare and gpt-image-2.5-sunburst. It is a step up from GPT Image 2: two extra quality levels (xhigh and max), custom sizes, transparent PNG and WebP output, and roughly half the latency, billed at the same $30 per million output tokens. Use Flare for most work and Sunburst for detailed layouts and precise edits. Its successor, GPT Image 2.6, arrived on 4 November 2026; 2.5 is still cheaper for drafts at low and medium quality and is the only one of the two with a half-price Batch API. On GPT Image Hub both 2.5 models cost 8 credits per image on the Saver route, at any size and quality.

## What changed from GPT Image 2

| Feature | GPT Image 2 | GPT Image 2.5 |
| --- | --- | --- |
| API models | gpt-image-2 | gpt-image-2.5-flare, gpt-image-2.5-sunburst |
| Quality levels | low, medium, high, auto | low, medium, high, xhigh, max, auto |
| Custom sizes | Fixed preset sizes | Multiples of 16, 1:3 to 3:1, no edge above 3840 px |
| Transparent background | Not supported | Yes, with PNG or WebP output |
| Reference images | Up to 16 | Up to 16 |
| Latency (OpenAI) | Baseline | About 50% lower |
| Image output price | $30 per 1M tokens | $30 per 1M tokens |

- xhigh and max are new quality levels above high. They use more output tokens, so keep them for approved compositions.
- Custom sizes let you ask for an exact WIDTHxHEIGHT instead of a preset. OpenAI labels sizes above 2560×1440 experimental.
- Transparent backgrounds work when the output format is PNG or WebP, which makes 2.5 useful for stickers, icons and cutouts.
- In our test on GPT Image Hub, a 1536×1024 image at high quality took 27 seconds on 2.5 Flare against 78 seconds on GPT Image 2, and used about a quarter of the output tokens.

GPT Image 2 is still a solid default, and on the Hub its Saver version is the cheapest OpenAI model at 5 credits. Our [step-by-step GPT Image 2 guide](https://www.gptimagehub.com/guides/how-to-use-gpt-image-2) covers it.

## GPT Image 2.5 Flare vs Sunburst

Both models take the same prompts, sizes, quality levels and reference images, and cost the same per token. OpenAI positions Flare as the fast default for everyday generation and Sunburst for workflows where editing precision matters most.

| Task | Pick | Why |
| --- | --- | --- |
| Social posts, product shots, posters | Flare | Fast, with clean text in our tests |
| Stickers, icons and cutouts | Flare | Transparent PNG or WebP at a lower wait |
| Packaging, labels and dense layouts | Sunburst | Placed every label exactly as prompted in our tests |
| Edits that must not touch the rest of the image | Sunburst | OpenAI's recommendation for precise edits |
| High-volume drafts | Flare | Sunburst takes longer per image |

In GPT Image Hub's tests Sunburst took 47 to 108 seconds per image depending on quality, against 27 to 60 seconds for Flare. On the same app-screen prompt, Flare added a word nobody asked for while Sunburst kept to the brief (images below).

## GPT Image 2.5 pricing

OpenAI bills GPT Image 2.5 at $30 per million image output tokens, the same rate as GPT Image 2, so the cost of an image depends on how many tokens its quality level uses. [CellCog](https://cellcog.ai/blog/gpt-image-2-6-vs-2-5/) measured a 1024×1024 image at each level:

| Quality level | Output tokens | Cost per image (API) |
| --- | --- | --- |
| low | 196 | $0.006 |
| medium | 439 | $0.013 |
| high | 1,758 | $0.053 |
| xhigh | 3,122 | $0.094 |
| max | 7,024 | $0.211 |

2.5 also has OpenAI's Batch API at half price, which GPT Image 2.6 does not have yet, so large overnight jobs are cheapest on 2.5.

On GPT Image Hub there are two ways to run it. The Saver versions cost a flat 8 credits per image (about $0.008–$0.011) at every size up to 4K and every quality level up to max. The standard GPT Image 2.5 Flare and Sunburst models are priced by quality and resolution:

| Quality level | 1K | 2K | 4K |
| --- | --- | --- | --- |
| low | 5 credits | 10 credits | 16 credits |
| medium | 11 credits | 22 credits | 36 credits |
| high | 43 credits | 86 credits | 143 credits |
| xhigh | 76 credits | 153 credits | 254 credits |
| max | 169 credits | 343 credits | 570 credits |

New accounts get 25 free credits, enough for 3 images on a 2.5 Saver model. See [GPT Image 2.5 Flare free](https://www.gptimagehub.com/gpt-image-2-5-flare-free) for what the free credits cover, and the [pricing page](https://www.gptimagehub.com/pricing) for credit packs.

## Real GPT Image 2.5 outputs

Every image below was generated on GPT Image Hub; captions give the size, quality and time.

### Flare

![Festival poster with the title FLARE made with GPT Image 2.5 Flare](https://www.gptimagehub.com/models/gpt-image-2-5-flare-poster.webp)
1024×1536, max quality, 50 s. Title rendered exactly as prompted.

![Six pastel coffee-cup stickers on a transparent background made with GPT Image 2.5 Flare](https://www.gptimagehub.com/models/gpt-image-2-5-flare-stickers.webp)
1024×1024, xhigh quality, transparent PNG, 60 s.

![Matte black speaker on a concrete plinth made with GPT Image 2.5 Flare](https://www.gptimagehub.com/models/gpt-image-2-5-flare-speaker.webp)
1536×1024, high quality, 27 s.

![The same speaker after an edit with GPT Image 2.5 Flare that only changed the grille to oxblood red](https://www.gptimagehub.com/models/gpt-image-2-5-flare-speaker-edit.webp)
Edit of the photo above: only the grille color changed, as instructed. 37 s.

### Sunburst

![Botanical cross-section of a fig with five hand-lettered captions made with GPT Image 2.5 Sunburst](https://www.gptimagehub.com/models/gpt-image-2-5-sunburst-fig.webp)
1024×1536, high quality, 47 s. All five captions placed and spelled correctly.

![Tea packaging with two label lines made with GPT Image 2.5 Sunburst](https://www.gptimagehub.com/models/gpt-image-2-5-sunburst-tea.webp)
1024×1024, xhigh quality, 73 s. Both label lines rendered as prompted.

![Illustrated seed packet made with GPT Image 2.5 Sunburst](https://www.gptimagehub.com/models/gpt-image-2-5-sunburst-seed-packet.webp)
1024×1536, max quality, 108 s.

### Same prompt on Flare and Sunburst

![Text-heavy event poster generated with GPT Image 2.5 Flare from the same prompt as the other models in this comparison](https://www.gptimagehub.com/guides/gpt-image-2-6/poster-2-5-flare.webp)
GPT Image 2.5 Flare · 1024x1536, quality high, 19.5 s. Title and date line spelled correctly.

![Text-heavy event poster generated with GPT Image 2.5 Sunburst from the same prompt as the other models in this comparison](https://www.gptimagehub.com/guides/gpt-image-2-6/poster-2-5-sunburst.webp)
GPT Image 2.5 Sunburst · 1024x1536, quality high, 17.7 s. Title and date line spelled correctly.

![Mobile app screen generated with GPT Image 2.5 Flare from the same prompt as the other models in this comparison](https://www.gptimagehub.com/guides/gpt-image-2-6/app-2-5-flare.webp)
GPT Image 2.5 Flare · 1024x1536, quality high, 19.9 s. Every requested string correct, plus an unrequested "Sunny".

![Mobile app screen generated with GPT Image 2.5 Sunburst from the same prompt as the other models in this comparison](https://www.gptimagehub.com/guides/gpt-image-2-6/app-2-5-sunburst.webp)
GPT Image 2.5 Sunburst · 1024x1536, quality high, 18.7 s. Every requested string correct and nothing added.

From our 5 November 2026 test of the 2.5 and 2.6 models: both spelled the poster's title and date line correctly; on the app screen Flare added an unrequested "Sunny" while Sunburst kept to the brief.

## GPT Image 2.5 prompts that work

Prompts written for GPT Image 2 carry over unchanged. Describe the subject, the setting, the light, any exact text in quotes, and the format; then add the 2.5-only parts, such as a transparent background or a custom size, explicitly. Our [GPT Image 2.6 prompt guide](https://www.gptimagehub.com/guides/how-to-prompt-gpt-image-2-6) has a six-line template that works on 2.5 too.

#### Product hero with headline space

1536×1024 · high · Flare

```text
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.
```

Try it: https://www.gptimagehub.com/generate?model=gpt-image-2-5-flare-saver&prompt=Create+a+1536x1024+product+hero+of+a+matte+black+wireless+speaker+on+a+pale+concrete+plinth%2C+soft+north-window+light%2C+shallow+depth+of+field%2C+blank+space+on+the+right+for+a+headline.

#### Transparent sticker sheet

1024×1024 · xhigh · PNG · Flare

```text
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.
```

Try it: https://www.gptimagehub.com/generate?model=gpt-image-2-5-flare-saver&prompt=Generate+a+transparent-background+PNG+sticker+set+of+six+hand-drawn+coffee+cups%2C+thick+outlines%2C+flat+pastel+fills%2C+no+shadows%2C+evenly+spaced+on+one+sheet.

#### Labeled illustration

1024×1536 · high · Sunburst

```text
Create a 1024x1536 editorial illustration of a botanical cross-section of a fig, labeled with five short hand-lettered captions, muted sage and plum palette, paper texture.
```

Try it: https://www.gptimagehub.com/generate?model=gpt-image-2-5-sunburst-saver&prompt=Create+a+1024x1536+editorial+illustration+of+a+botanical+cross-section+of+a+fig%2C+labeled+with+five+short+hand-lettered+captions%2C+muted+sage+and+plum+palette%2C+paper+texture.

## GPT Image 2.5 vs 2.6

GPT Image 2.6 (4 November 2026) keeps the Flare and Sunburst split and the token price. From high quality upward it uses the same tokens as 2.5, so it costs the same, and 2.6 Flare was about twice as fast in an independent API test; it also adds animated GIF and SVG output. At low and medium quality 2.6 uses 3 to 4 times more tokens, and it has no Batch API yet, so 2.5 remains the cheaper choice for drafts and bulk jobs. The full comparison, with side-by-side images, is in [GPT Image 2.6 vs 2.5](https://www.gptimagehub.com/guides/gpt-image-2-6-vs-2-5).

## How to try GPT Image 2.5 on GPT Image Hub

- Open the [generator with GPT Image 2.5 Flare Saver](https://www.gptimagehub.com/generate?model=gpt-image-2-5-flare-saver) selected, or switch to Sunburst Saver in the model picker.
- Write your prompt, pick an aspect ratio and a resolution (1K, 2K or 4K), and attach up to 16 reference images if you want to edit or match a style.
- Generate. Each image costs 8 credits on either Saver model, whatever the size or quality.
- New accounts start with 25 free credits; more are on the [pricing page](https://www.gptimagehub.com/pricing).

## GPT Image 2.5 FAQ

### What is GPT Image 2.5?

OpenAI's image generation model released on 8 September 2026, available as gpt-image-2.5-flare (fast, everyday) and gpt-image-2.5-sunburst (precision). It adds xhigh and max quality, custom sizes and transparent backgrounds over GPT Image 2.

### When was GPT Image 2.5 released?

On 8 September 2026, according to OpenAI's API changelog; OpenAI announced it as ChatGPT Images 2.5. GPT Image 2.6 followed on 4 November 2026.

### Is GPT Image 2.5 better than GPT Image 2?

For most work, yes: OpenAI describes it as higher quality at about half the latency, it adds xhigh and max quality and transparent backgrounds, and in our tests it used far fewer output tokens at high quality. GPT Image 2 is still a reliable, cheaper default on GPT Image Hub's Saver route.

### What is the difference between GPT Image 2.5 Flare and Sunburst?

Flare is faster and is OpenAI's default for everyday generation. Sunburst takes longer and is recommended when precision matters most, such as packaging, labels, dense layouts and careful edits. Both take the same settings and cost the same per token.

### How much does GPT Image 2.5 cost?

In OpenAI's API, $30 per million output tokens: about $0.006 for a 1024×1024 image at low quality, $0.053 at high and $0.211 at max, with half-price Batch. On GPT Image Hub the Saver versions cost 8 credits per image at any size and quality.

### Can I use GPT Image 2.5 for free?

New GPT Image Hub accounts get 25 free credits, which cover 3 images on a GPT Image 2.5 Saver model. After that, images are paid with credits.

### Should I use GPT Image 2.5 or 2.6?

Use 2.6 Flare for final images at high quality or above: same cost, faster, and it can make GIFs and SVGs through the API. Use 2.5 for cheap drafts at low or medium quality and for Batch jobs. See [GPT Image 2.6 vs 2.5](https://www.gptimagehub.com/guides/gpt-image-2-6-vs-2-5).

## Sources

Specifications are from OpenAI's documentation and announcement. API token counts are from CellCog's independent test. Images, timings and Hub prices are our own, from GPT Image Hub.

- [OpenAI API changelog (GPT Image 2.5, 8 September 2026)](https://developers.openai.com/api/docs/changelog)
- [OpenAI: Introducing ChatGPT Images 2.5](https://openai.com/index/introducing-chatgpt-images-2-5/)
- [OpenAI image generation guide](https://developers.openai.com/api/docs/guides/image-generation)
- [CellCog: GPT Image 2.6 vs 2.5, 80 generations on launch night](https://cellcog.ai/blog/gpt-image-2-6-vs-2-5/)

## Keep exploring

- [GPT Image 2.5 Flare model guide](https://www.gptimagehub.com/models/gpt-image-2-5-flare)
- [GPT Image 2.5 Sunburst model guide](https://www.gptimagehub.com/models/gpt-image-2-5-sunburst)
- [GPT Image 2.5 Flare free](https://www.gptimagehub.com/gpt-image-2-5-flare-free)
- [GPT Image 2.5 Sunburst free](https://www.gptimagehub.com/gpt-image-2-5-sunburst-free)
- [GPT Image 2.6 vs 2.5](https://www.gptimagehub.com/guides/gpt-image-2-6-vs-2-5)
- [GPT Image 2.6 guide](https://www.gptimagehub.com/guides/gpt-image-2-6)
- [How to use GPT Image 2](https://www.gptimagehub.com/guides/how-to-use-gpt-image-2)