OpenAI GPT Image 2.5 Flare Image edit pricing
openai/gpt-image-2.5-flare/edit
GPT Image 2.5 Flare, the tier OpenAI describes as fast and balanced. Five quality steps times three resolution tiers gives it fifteen prices, the largest grid on this page, shared with Sunburst.
What it costs
Per image, and what that comes to at volume. Every figure is computed through the same function the app estimates a real run with.
| Tier | per image | 1 image | 10 images | 100 images | 1,000 images |
|---|---|---|---|---|---|
| low, 1k | $0.025 | $0.025 | $0.25 | $2.50 | $25.00 |
| low, 2k | $0.035 | $0.035 | $0.35 | $3.50 | $35.00 |
| low, 4k | $0.045 | $0.045 | $0.45 | $4.50 | $45.00 |
| medium, 1k | $0.039 | $0.039 | $0.39 | $3.90 | $39.00 |
| medium, 2k | $0.055 | $0.055 | $0.55 | $5.50 | $55.00 |
| medium, 4k | $0.085 | $0.085 | $0.85 | $8.50 | $85.00 |
| high, 1k | $0.105 | $0.105 | $1.05 | $10.50 | $105.00 |
| high, 2k | $0.165 | $0.165 | $1.65 | $16.50 | $165.00 |
| high, 4k | $0.285 | $0.285 | $2.85 | $28.50 | $285.00 |
| xhigh, 1k | $0.175 | $0.175 | $1.75 | $17.50 | $175.00 |
| xhigh, 2k | $0.285 | $0.285 | $2.85 | $28.50 | $285.00 |
| xhigh, 4k | $0.495 | $0.495 | $4.95 | $49.50 | $495.00 |
| max, 1k | $0.375 | $0.375 | $3.75 | $37.50 | $375.00 |
| max, 2k | $0.615 | $0.615 | $6.15 | $61.50 | $615.00 |
| max, 4k | $1.015 | $1.015 | $10.15 | $101.50 | $1,015.00 |
- Sizing
- aspect_ratio plus resolution
- Max input images
- 16
Things that catch people out
- Bills $0.015 extra for every input image beyond the first.
- Accepts up to 16 input images, the highest ceiling here, shared with GPT Image 2 and Sunburst.
- Quality runs low, medium, high, xhigh and max, two steps past GPT Image 2.
What it accepts
Reference image
Takes input images, which is what holds a persona's face consistent between runs.
Aspect ratio
15 ratios.
Resolution tiers
3 tiers, and they are priced differently.
Quality control
5 steps (low, medium, high, xhigh, max), each priced differently.
Web search
Not offered on this endpoint.
Image search
Not offered on this endpoint.
Aspect ratios (15)
Others in this lane
Image to image. You supply one or more reference images and a prompt, and the model returns a new image of the same subject. This is the lane every persona workflow runs on, because the reference is what holds the face steady.