GPT Image 2.6 Flare and Sunburst
GPT Image is OpenAI’s image generation and editing family. GPT Image 2.6 Flare and Sunburst arrived on November 4, 2026 with the same token rates as 2.5. Flare is positioned as the default for most applications; Sunburst focuses on intricate detail with longer generation times. Our 2.6 pages include one real animated GIF per variant, with exact settings and billed cost. The common prompt benchmarks and measured comparisons still belong to the earlier versions.
OpenAI · facts checked
Which one to pick

GPT Image 2.6 Flare
$0.053 per 1024 by 1024, high quality via API list price
Clean text, transparency, careful edits and quick GIFs

GPT Image 2.6 Sunburst
$0.053 per 1024 by 1024, high quality via API list price
Intricate hero art, showcase GIFs and demanding revisions, slower

GPT Image 2.5 Flare
$0.053 per 1024 by 1024, high quality via fal
Existing projects tuned to 2.5, or inspecting our recorded Flare samples.

GPT Image 2.5 Sunburst
$0.053 per 1024 by 1024, high quality via fal
Existing Sunburst projects and our recorded 2.5 edit comparisons.

GPT Image 2
$0.21 per 1024 by 1024, high quality via fal
Earlier workflows that depend on GPT Image 2.
Side by side, measured
| Model | Price | Seconds per image | Cost of its fixed set | Sizes |
|---|---|---|---|---|
| GPT Image 2.6 Flare | $0.053 per 1024 by 1024, high quality via API list price | Not run | ||
| GPT Image 2.6 Sunburst | $0.053 per 1024 by 1024, high quality via API list price | Not run | ||
| GPT Image 2.5 Flare | $0.053 per 1024 by 1024, high quality via fal | 16 to 66 | $0.54 12 prompts | Exact pixel dimensions from a 1:3 strip to a 3:1 banner, up to 3840 on the long edge |
| GPT Image 2.5 Sunburst | $0.053 per 1024 by 1024, high quality via fal | 23 to 41 | $0.54 12 prompts | Exact pixel dimensions from a 1:3 strip to a 3:1 banner, up to 3840 on the long edge |
| GPT Image 2 | $0.21 per 1024 by 1024, high quality via fal | 72 to 163 | $2.16 12 prompts | Ask for exact pixel dimensions, anything from a 1:3 strip to a 3:1 banner up to 3840 on the long edge |
Seconds and cost are from our own runs of each model's fixed set: the same 12 prompts for every model that makes pictures from a prompt. Open a model for every picture and the request behind it.
Every version
| Version | Released | What changed |
|---|---|---|
| GPT Image 1 gpt-image-1 | First API release of the model behind ChatGPT's March 2025 image generation. Text rendering, edits and transparent backgrounds. | |
| GPT Image 1 Mini gpt-image-1-mini | Cost-cut variant, roughly 80 percent cheaper than GPT Image 1. | |
| GPT Image 1.5 gpt-image-1.5 | Precise edits that keep untouched detail intact, up to four times faster, 20 percent cheaper. | |
| GPT Image 2 gpt-image-2 | Reasoning before drawing, custom pixel sizes up to 3840 on the long edge, three quality tiers, batch support. | |
| GPT Image 2.5 Flare gpt-image-2.5-flare | The fast default. Up to 50 percent lower latency than Images 2.0, better lighting and textures, xhigh and max quality tiers. | |
| GPT Image 2.5 Sunburst gpt-image-2.5-sunburst | The precision tier, slower, for edits where fidelity matters most. Same token prices as Flare. | |
| GPT Image 2.6 Flare gpt-image-2.6-flare | Newer generation and editing model at the same token rates as 2.5. Quality and reasoning effort are separate settings. | |
| GPT Image 2.6 Sunburst gpt-image-2.6-sunburst | Newer generation and editing model at the same token rates as 2.5. Quality and reasoning effort are separate settings. |
Frequently asked questions
Does the 2.6 API release establish which model ChatGPT uses?
No. These pages document the 2.6 generation and editing endpoints. They do not verify the model version selected inside ChatGPT.
Is GPT Image the same as DALL-E?
No. DALL-E was OpenAI's earlier image family. GPT Image replaced it in 2025 with a model that generates images inside the GPT model itself, which is why it handles text and instructions differently.
Do the GPT Image models have a free tier?
In ChatGPT, yes, with an unpublished daily cap. On the API, no.