Three GPT Image API models are scheduled to retire December 1, 2026
OpenAI schedules gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest for API removal on December 1, 2026. Its current deprecation table recommends gpt-image-2.5-sunburst or gpt-image-2.5-flare.
Required action
Inventory the three retiring IDs and test the currently named Sunburst or Flare replacement before December 1. Retain the earlier GPT Image 2 guidance in correction history; do not transfer version-specific settings.
Scope and evidence limits
OpenAI API; no regional exception documented
The removal applies to the OpenAI API; the official notice lists no regional exception.
not_tested
Official sources
- developers.openai.com · changelog
- developers.openai.com · deprecations
- developers.openai.com · gpt-image-2
Affected components
gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest
Affected users
API developers calling the listed GPT Image model IDs.
Migration guidance
Evaluate gpt-image-2.5-sunburst or gpt-image-2.5-flare for the affected workflow. Their September 8 launch documents Images API and Responses image-generation tool support. Validate the exact model, endpoint, parameters, outputs and costs before migration.
Correction history
- 2026-09-12: Rechecked the three exact model IDs and December 1 deadline. Replaced premature Retired status with scheduled removal, clarified the exact replacement endpoint scope, and removed an unsupported no-migration-fee claim.
- 2026-09-21: The phrase "not Responses" conflates direct model endpoint support with the Responses image-generation tool.
- 2026-10-01 source review: Current HTML and Markdown tables now name Image 2.5 Sunburst/Flare rather than GPT Image 2; exact retiring IDs and December 1 remain unchanged. Previous guidance: OpenAI schedules gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest for API removal on December 1, 2026. Its deprecation table names gpt-image-2 as the replacement. Action: Inventory the three exact retiring IDs. Test gpt-image-2 on its supported endpoint and parameters, then compare representative generation, editing and cost before December 1.