Both GPT Image 2.6 models plan at high complexity unless you say otherwise. For small edits like swapping a background, set complexity to low, and save high for dense text and complex layouts.
OpenAI's new GPT Image 2.6 family comes in two models, gpt-image-2.6-sunburst and gpt-image-2.6-flare, and both accept a new complexity request field. It controls how much the model plans before it renders, and the supported levels are low, medium and high. The detail most teams will miss is the default: if you omit complexity, both models use high.
That default suits hard jobs and works against easy ones. OpenAI's image generation guide says higher complexity can help with dense text, complex layouts and relationships between objects, but it can also increase latency and introduce unwanted changes. The same guide says lower complexity can suit focused edits, such as changing a background or a sign. In other words, if your pipeline sends a quick touch up at the default setting, you are asking the model to replan an image you only wanted nudged, and you may get back more change than you asked for.
Choosing between the two models is simpler than it looks, because OpenAI lists the same token prices for both: $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. Price is not the deciding factor. OpenAI positions Sunburst for detailed visual work and Flare for fast iteration, and says both support markup guided edits, SVG output and animated GIF output through the Image API.
The practical pattern is to treat complexity as a per request decision rather than a global setting. Route quick edits and drafts through Flare at low complexity, and send finished assets with heavy text or intricate layouts to Sunburst at high. Because the right level depends on your images, run the same prompts at each level and compare the results before you lock in a default.
Try it today
- Find every GPT Image 2.6 call in your code that omits complexity. Those requests are running at high on both Flare and Sunburst.
- Set complexity to low for focused edits such as swapping a background or changing the text on a sign.
- Keep complexity at high only for images with dense text, complex layouts or many objects that must relate correctly.
- Use Flare for fast iteration and Sunburst for detailed final work. The listed token prices are the same, so choose on speed and fit, not cost.
- Run a small side by side test of low, medium and high on your own prompts, check latency and unwanted changes, then set a default per workflow.