> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kymaapi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# GPT Image 2.5 Flare

> GPT Image 2.5 Flare is OpenAI’s September 2026 image model, the variant built for fast everyday generation. Reach for it when a character or a face has to stay the same across several images, or when the text inside the image has to be exact.

## Overview

Released in September 2026, GPT Image 2.5 comes in two variants; Flare is the one tuned for everyday generation. It accepts a text prompt plus optional reference images and returns images at 1024×1024, 1024×1536 or 1536×1024, with low, medium and high quality tiers (medium is the default). Each reference image adds a small charge once per request.

In our own tests it held a person’s identity across reference images, rendered logo and slide text letter-exact, and finished a medium image in roughly 10 seconds. Billing is per image at the chosen quality tier, and the response carries the exact cost in usage.cost.

It generates images only: no text output, structured outputs or reasoning controls. It is not a vector model, so use a vector SKU when you need SVG.

## Specs

| Field | Value |
| - | - |
| Model ID | `gpt-image-2.5-flare` |
| Best for | Consistent characters from references, text-in-image, portraits |
| Input modalities | Text, Image |
| Output modalities | Image |
| Speed | Fast |
| Cost band | Balanced |
| Release stage | Stable |

## Pricing

| | Price |
| - | :-: |
| Per image | \$0.019 |

Billed per generated image. Every request reports its exact cost in `usage.cost`; the full media price list is on [Pricing](/pricing).

## Use this when

* **Consistent Characters** — Keeps the same face, hair and details across a series of images from one or more reference photos.
* **Text-Exact Graphics** — Renders wordmarks, slide titles and social cards with the lettering spelled as written.
* **Portraits From References** — Produces photoreal portraits and scenes of a real subject from a handful of reference images.
* **Reference Image Editing** — Changes setting, lighting or outfit while preserving the subject in the reference.

## Not ideal for

Do not use this model for vector output, structured JSON, or the cheapest possible bulk drafts.

## Pick something else when

* You need the previous generation with the same tiers: use [`gpt-image-2`](/models/gpt-image-2).
* You require scalable vector graphics: use [`recraft-v4-vector`](/models/recraft-v4-vector).
* You need sub-cent bulk drafts: use [`minimax-image-01`](/models/minimax-image-01).

## Example

```python theme={null}
from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="kyma-...")

response = client.images.generate(
    model="gpt-image-2.5-flare",
    prompt="...",
)
```

## Agent query example

Ask the API which models fit, instead of hardcoding an id:

```bash theme={null}
curl "https://kymaapi.com/v1/models?recommended_for=generate-images&quality_tier=frontier-open"
```

## FAQ

**Is this the same as gpt-image-2.5?**

Yes. GPT Image 2.5 is published as two variants, Flare and Sunburst; Kyma serves Flare under the id gpt-image-2.5-flare.

**Which quality tiers can I request?**

low, medium and high. Medium is the default, and the price follows the tier you pick.

**Can I pass reference images?**

Yes. Send image\_url or image\_urls (up to 16) with the request and the model edits from those references. Each reference adds a small charge once per request.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.