Model comparison
GPT Image 2.6 Flare vs Sunburst
GPT Image 2.6 ships in the API as two models. Flare is built for speed — drafts in as little as six seconds. Sunburst is built for quality and keeps characters, products and brand details consistent across edits. Here is how to choose.
Side-by-Side Comparison
| Feature | Flare | Sunburst |
|---|---|---|
| Tuned for | Low latency | Detail and consistency |
| Speed | As fast as ~6 s at low quality | Slower, scales with quality |
| Recommended quality | Low or medium | Medium or high |
| Edit consistency | Good | Best — characters and products hold across edits |
| Layout for posters & product shots | Improved over 2.5 | Strong |
| API list price (per 1M tokens) | Text in $5 · image in $8 · image out $30 | Same token rates as Flare |
| Approx. cost per image | ~$0.005 (1K, low) | ~$0.05 (1K, high) · ~$0.10 (2K, high) |
| Credits on gptimage26 | 1 per 1K image | 4 per 1K · 8 per 2K |
Prices are OpenAI list rates as reported at launch; check OpenAI's pricing page for current numbers.
When to Use Flare
- Exploring many ideas or styles quickly
- Social posts, thumbnails and memes
- Testing GIF motion before a final render
- High-volume jobs where cost matters most
When to Use Sunburst
- Final product photos and campaign visuals
- Posters and layouts with readable text
- Multi-step edits of the same character or product
- 2K output for print and large screens
Our recommendation
Draft with Flare at low quality until the composition is right, then regenerate the winner with Sunburst at medium or high quality. You get Flare's speed for iteration and Sunburst's detail where it counts.
Try both modelsFrequently Asked Questions
Not always. Sunburst gives more detail and better consistency across edits, but Flare is much faster and cheaper. For exploring ideas, Flare usually wins; for the final asset, switch to Sunburst.
Third-party reports list the same per-token rates for both models. The real difference comes from the quality and size you choose, which change how many output tokens an image uses.
Start with Flare to test the motion quickly. When the loop works, regenerate it with Sunburst for cleaner frames and a more consistent character.
Complexity (low, medium, high) controls how much planning the model does before drawing. OpenAI recommends low complexity for quick edits and medium or high for crowded scenes with many constraints.