Start with Flare for everyday image generation. Consider Sunburst when precise editing is your priority. That is OpenAI's positioning, not the result of our own head-to-head test. If GPT Image 2 already works for you, test your usual prompts before switching.
What is different about Sunburst and Flare?
| Model | OpenAI's positioning | What to check on your own work |
|---|---|---|
| GPT Image 2.5 Flare | Fast, high-quality everyday generation | Does it produce a usable image faster? |
| GPT Image 2.5 Sunburst | Capable generation and precise editing | Does it make the requested change while preserving the rest? |
| GPT Image 2 | Earlier generation model; our existing comparison is separate | Is the improvement worth changing your workflow? |
The exact 2.5 model IDs are gpt-image-2.5-flare and
gpt-image-2.5-sunburst. Both accept text and image inputs. Both support
low, medium, high, xhigh,
max, and auto quality settings.
A model described as more capable is not automatically the best choice for every image. A quick product draft and a careful edit to an approved design are different jobs.
Does GPT Image 2.5 cost more per image?
The token rates match GPT Image 2, but the cost of an image can differ. Both 2.5 models list these standard API rates in US dollars:
| Billed item | Price per million tokens |
|---|---|
| Text input | $5 |
| Image input | $8 |
| Image output | $30 |
These are token prices, not a fixed price for each image. The final bill depends on the tokens used by the prompt, any reference images, and the generated output. Cached input has separate discounted rates.
Do not carry over the old GPT Image 2 "$0.053 per medium image" estimate to 2.5. OpenAI explicitly says the GPT Image 2 calculator does not estimate GPT Image 2.5 token consumption. These API charges are also different from ChatGPT subscription allowances.
What can a generated product image look like?
Submitted prompt: Create a studio-quality product photo of a premium sparkling water can called "AURORA" on a wet black stone surface with lemon slices and soft rim lighting. The can label should clearly read "AURORA", "Sparkling Water", and "Lemon". Photorealistic, commercial ad style, high detail.
The requested label text is readable in this example. One successful image does not establish how consistently the model follows prompts or how it compares with Sunburst or Image 2.
How can you compare it with GPT Image 2?
Use the same prompt, output size and requested quality for each model. Our existing GPT Image 2 comparison includes three prompts you can reuse:
- A sparkling-water product photo with readable label text.
- A Kyoto travel advertisement with a headline, price and booking text.
- A water-cycle infographic with four labels and arrows.
Check the finished images, time each request, and use returned token usage to calculate the cost. We have not run this matched three-model benchmark; the older Image 2 images are not Image 2.5 results.
Should you switch now?
For now, use the official positioning as a shortlist, not a guarantee. Try Flare for a generation-heavy workflow and evaluate Sunburst separately for editing. Keep GPT Image 2 if it already meets your needs until a matched test shows a useful improvement.
For an upgrade decision, ask three simple questions: Is the result usable? How long did it take? What did that usable result cost, including retries?
Sources
- OpenAI GPT Image 2.5 Flare: https://developers.openai.com/api/docs/models/gpt-image-2.5-flare
- OpenAI GPT Image 2.5 Sunburst: https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst
- OpenAI GPT Image 2: https://developers.openai.com/api/docs/models/gpt-image-2