GPT Image 2.5 Sunburst and Flare: choose around the image job

The release

OpenAI introduced gpt-image-2.5-sunburst and gpt-image-2.5-flare on September 8, 2026. The release positions Sunburst for precise edits and Flare for faster everyday generation.

Separate precision from throughput

For a detailed edit, define the parts of the source that must remain unchanged. For a generation queue, define the acceptable quality and delivery window. These are different acceptance criteria and deserve separate pilot results.

Use the exact API identifiers in configuration and reporting. A family name alone makes it difficult to tell which choice produced a particular asset.

Run a useful comparison

Choose a small set of real briefs with known review requirements. Record candidate spend, completed requests, reviewer corrections, and approved outputs. Keep the first comparison at consistent settings before exploring additional quality options.

The provider’s positioning is a starting hypothesis for that test, not a measured result for your application. Do not publish a speed or quality claim based solely on the model name.

Adopt only the choices your team needs

If one option is sufficient for a workflow, keep its default simple. If both are useful, label the presets by task and make the model ID visible to developers. Review current access and rates in the XMH.NET console before enabling a new production preset.