OpenAI is deprecating three GPT Image generation models — gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest — and removing them from the API, directing developers to gpt-image-2

OpenAI is retiring the older models in its GPT Image line. The company’s API deprecations page records that on 2 June 2026 it notified developers using those models of their deprecation and removal from the API on 1 December 2026. Three identifiers are affected: gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest. All three are directed to a single replacement, gpt-image-2.

This is a gentler shape of sunset than OpenAI’s Sora 2 removal earlier in the year. There, programmatic video generation was leaving the API with no in-platform substitute, forcing a provider decision. Here, image generation stays — only the specific model versions change. OpenAI names gpt-image-2 as the successor for each of the three, so for most integrations the work is updating a model string rather than rebuilding around a different service.

The one identifier worth a second look is chatgpt-image-latest. The word “latest” reads like a floating alias that would follow OpenAI to the new model automatically, but it appears in the deprecation table as a model being shut down on 1 December 2026, pointed at gpt-image-2 like the other two. Treat it as a fixed name being removed, not one that will quietly roll forward — code pinned to it needs an explicit change.

Deprecated models keep responding until the date, so nothing breaks immediately. The exposure is any production path that names one of the three: on-demand image generation or editing, an asset pipeline, a feature built on the mini model for cost. The safe move is to swap in gpt-image-2 ahead of 1 December 2026 and verify output quality, parameters, and cost against what you depend on now — image models are not always drop-in identical across versions — rather than discovering the difference when the old strings stop answering.