OpenAI's gpt-image-2.5 ships as two models. Flare is the fast one, with about 50% lower latency than GPT Image 2. Sunburst is the base model tuned for precision. Generate and edit online, without watermark, at 4 credits per image on Aura AI.
GPT Image 2.5 is OpenAI's image generation and editing release of September 8, 2026, the successor to GPT Image 2 from April 2026. GPT Image 2.5 Flare (API id gpt-image-2.5-flare) is a small model optimized for speed: OpenAI rates its image quality as comparable to GPT Image 2 while cutting latency by about 50%. GPT Image 2.5 Sunburst (gpt-image-2.5-sunburst) is the base model optimized for quality, with higher image quality than GPT Image 2 and longer generation times, built for detailed creative work.
Both share the same generation of improvements: improved fidelity for more natural, recognizable images, better subject preservation in edits, consistent details across a chain of edits, and comment-based edits that change only what you ask for. In the API, OpenAI added two quality tiers above high, xhigh and max, an opaque background option, and custom sizes with a long edge up to 3,840 pixels, while keeping the token cost per image identical to GPT Image 2. At launch, Arena.ai reported Sunburst in first place and Flare in second across its text-to-image and image-edit leaderboards, ahead of GPT Image 2, on preliminary scores.
On Aura AI, GPT Image 2.5 runs in the text-to-image tool as Flare and Sunburst, each at a flat 4 credits per image, with size presets and several images per request. In the image editor, Flare Edit and Sunburst Edit take one input image and cost 4 credits per edit. Output is delivered without watermark, and the same credits work on Nano Banana 2, Seedream 5.0, FLUX 2 and the other 40+ models in the catalog.
Same price on Aura AI, different trade-off between speed and precision
| Feature | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst |
|---|---|---|
| Model type | Small model optimized for speed | Base model optimized for quality |
| Speed | About 50% lower latency than GPT Image 2 | Longer generation times |
| Precision | Quality comparable to GPT Image 2, with the 2.5 editing gains | Higher quality than GPT Image 2, holds intricate detail |
| Best for | Drafts, iterations, social visuals, batches of variations | Final assets, product and packaging shots, detailed illustration |
| Arena rank at launch (reported) | 2nd | 1st |
| Credits on Aura AI | 4 per image, 4 per edit | 4 per image, 4 per edit |
What the model does upstream and how it is exposed on Aura AI
| Feature | GPT Image 2.5 on Aura AI |
|---|---|
| Modes | Text to image (Flare, Sunburst), image edit (Flare Edit, Sunburst Edit) |
| Input images for edits | 1 per request |
| Output sizing | Image size presets, chosen before generation |
| Images per request | Multiple images in one run |
| Text in images | Quoted wording rendered where the prompt places it |
| Editing behavior | Subject preservation, consistent details across edits, comment-based edits |
| Quality levels (upstream API) | auto, low, medium, high, xhigh, max |
| Custom sizes (upstream API) | Long edge up to 3,840 px, edges in multiples of 16, ratio up to 3:1 |
| Background (upstream API) | auto, opaque, transparent |
| Watermark | None |
| Plan | Starter, Pro, Premium |
Six changes that separate gpt-image-2.5 from the April 2026 model
GPT Image 2 was a single model. GPT Image 2.5 splits into Flare, the small model for speed, and Sunburst, the base model for quality, so you choose the trade-off per image.
Flare delivers image quality comparable to GPT Image 2 with about 50% lower latency. That halves the wait between a prompt and the next revision.
OpenAI describes the 2.5 generation as producing more natural, recognizable images, and Sunburst as higher quality than GPT Image 2 outright.
Chain several edits and the parts you did not touch stay put. Pass the previous output as the next edit input and subject, lighting and layout carry over.
Point at one thing and say what should change. GPT Image 2.5 applies the note to that part only and leaves the rest untouched.
In the API, two quality levels sit above high, xhigh and max, and background gains an opaque setting alongside transparent. The token cost per image stays identical to GPT Image 2.
Four steps from prompt to image, or from a photo to a precise edit
Flat cost per image, no size multiplier
| Variant | Mode | Credits per image | Best for |
|---|---|---|---|
| GPT Image 2.5 Flare | Text to image | 4 credits | Fast drafts, iterations, variations |
| GPT Image 2.5 Flare Edit | Image edit, 1 input | 4 credits | Quick precise edits with subject preservation |
| GPT Image 2.5 Sunburst | Text to image | 4 credits | Final assets, detailed creative work |
| GPT Image 2.5 Sunburst Edit | Image edit, 1 input | 4 credits | Edits where intricate detail must hold |
| GPT Image 2 | Text to image or edit | 4 credits | The April 2026 model, for matching earlier work |
| GPT Image 1.5 | Text to image or edit | 2 credits | Lowest-cost OpenAI drafts |
How the new OpenAI pair compares to its predecessor and the other flagship image models on Aura AI
| Feature | GPT Image 2.5 | GPT Image 2 | Nano Banana 2 | Seedream 5.0 |
|---|---|---|---|---|
| Maker | OpenAI | OpenAI | Google DeepMind | ByteDance |
| Release | Sep 2026 | Apr 2026 | Feb 2026 | Jul 2026 (5.0 Pro) |
| Variants | Flare (speed), Sunburst (quality) | Single model | Nano Banana 2, Pro | 5.0, 5.0 Pro |
| Credits, text to image | 4 | 4 | 3 | 3 |
| Credits, image edit | 4 | 4 | 3 | 2 to 4 |
| Fast variant | ✓ Flare, about 50% lower latency | No | Not specified | Not specified |
| Max resolution (upstream) | 3,840 px long edge | 3840 x 2160 | 4K | 4K |
| Plan | All plans | All plans | All plans | All plans |
Common questions about Flare and Sunburst on Aura AI
Flare for speed, Sunburst for precision, 4 credits per image, no watermark. Available on every plan, worldwide.