GPT-Image-2.6 (Flare and Sunburst)
modelYour notes
OpenAI's image models released to ChatGPT, Codex and the API on November 4, 2026, eight weeks after GPT-Image-2.5. It keeps the same two tiers: Sunburst (gpt-image-2.6-sunburst) for detailed compositions and editing, and Flare (gpt-image-2.6-flare) for fast iteration. The system card says both improve editing consistency, detail preservation and responses to sketches and image annotations. The release adds three features. A complexity setting (low, medium or high, with high as the default) controls how much the model plans before rendering, separately from the quality setting. It also adds animated GIF generation, markup-guided edits (also available on 2.5) and, in the API, SVG output. Both tiers carry the same token prices as 2.5 Sunburst: $5 per million text-input tokens, $8 image input and $30 image output. Architecture and parameters are undisclosed.
The system card describes a safety stack carried over from Images 2.5, with added safeguards for GIF creation. On 3,077 adversarial prompts, unsafe outputs that moderation failed to block fell from 1.92% to 1.72% for Sunburst and rose from 1.49% to 2.14% for Flare against the matching 2.5 baselines; neither change in that measure is marked statistically significant.
Model Details
Variants
| Name | Parameters | Notes |
|---|---|---|
| GPT-Image-2.6 Sunburst | — | Detailed generation and editing; $5/$8/$30 per M text-in/image-in/image-out |
| GPT-Image-2.6 Flare | — | Fast iteration; same token prices as Sunburst |