SunsettingAI Image

gpt-image-1, gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest

by OpenAI

Announced

2026-04-22

Complete

2026-12-01

Refund window closes

—

Refund status

No refund applicable

The whole GPT Image 1 family loses API access in two waves. OpenAI's deprecations page puts `gpt-image-1` in the April 22, 2026 legacy-snapshot batch with a shutdown date of October 23, 2026, and `gpt-image-1-mini`, `gpt-image-1.5` and `chatgpt-image-latest` in a separate June 2, 2026 announcement with a shutdown date of December 1, 2026. All four name the same substitutes: `gpt-image-2.5-sunburst` or `gpt-image-2.5-flare`. Two things change beyond the model string, and both were read from OpenAI's own model pages on September 29, 2026. First, all four retiring models list the Batch API as supported and neither replacement does, so a bulk still-frame or thumbnail job running through `v1/batch` has no documented replacement path. Second, the four retiring models publish flat per-image prices by quality and size, and the two replacements publish token pricing only, so a per-image budget has to be rebuilt as a token estimate.

Refund flow

1
Not applicable. These are pay-per-image and pay-per-token API models with no pre-paid balance tied to a model string, so there is no refund category to open.
2
If you are billed for `gpt-image-1` after October 23, 2026, or for `gpt-image-1-mini`, `gpt-image-1.5` or `chatgpt-image-latest` after December 1, 2026, treat it as a billing error rather than a refund request: open a ticket at help.openai.com with the invoice line and the request IDs.
3
The number worth checking before either date is not a refund, it is your per-image cost after the switch. The retiring models are priced per image; the replacements are priced per token, and OpenAI states on both replacement model pages that its GPT Image 2 calculator does not estimate GPT Image 2.5 token consumption, so the estimate has to come from your own metered test run.

Migration path

Recommended successor

gpt-image-2.5-sunburst or gpt-image-2.5-flare

Why this is the closest fit

They are the substitutes OpenAI names itself, on every row of both deprecation batches, and the two models its current image generation guide tells you to pick between: Sunburst where editing precision matters most, Flare for fast everyday generation.

What differs from the original

Three differences show up in the model pages rather than the announcement. The Batch API is the sharp one: `gpt-image-1`, `gpt-image-1-mini`, `gpt-image-1.5` and `chatgpt-image-latest` all list `v1/batch` as supported, and `gpt-image-2.5-sunburst` and `gpt-image-2.5-flare` both list it as not supported, so bulk overnight generation has to move to live calls with your own rate-limit handling. The Responses API changes shape too: `gpt-image-1` is the one retiring model that supports `v1/responses` directly, while the 2.5 models do not, and the guide instead has you pick a mainline model and name the image model in the image generation tool's `model` field. Pricing changes unit: the retiring models list flat per-image prices by quality and size, while the replacements list token rates only, at $8 per million image input tokens and $30 per million image output tokens. Quality settings are wider on the replacements, which accept low, medium, high, xhigh, max and auto. Inpainting is listed as a supported feature on both generations, so masked edits survive the move.

Alternatives by use case

You run bulk image jobs through the Batch API

There is no documented like-for-like path. gpt-image-2 still lists Batch as supported, so check whether it fits before rewriting your queue against live rate-limited calls, and re-read the endpoint table on the model page before you commit because it can change

You call gpt-image-1-mini for cheap draft frames

This is the price point that disappears. gpt-image-1-mini lists low-quality 1024x1024 at $0.005 an image; the replacements quote tokens, not images, so meter a representative batch and derive your own per-image number before switching

You generate a still and feed it to an image-to-video model

Move the still step first and re-run your reference frames, because a new image model changes composition and text rendering, and both of those propagate into every frame the video model produces from that still

You call gpt-image-1 as a model inside the Responses API

Switch to a mainline model with the image generation tool and set the tool model to gpt-image-2.5-sunburst or gpt-image-2.5-flare, since the 2.5 models are not listed as supported on v1/responses directly

You pinned chatgpt-image-latest to match what ChatGPT produced

OpenAI's own model page for it now recommends gpt-image-2.5-sunburst for API use, so take that path and re-baseline your prompts rather than hunting for another ChatGPT-matching alias

What this tool meant

GPT Image 1 is the model that made a generated still a routine step in a video pipeline rather than a separate project. Between its launch and these shutdown dates, the normal way to build an AI video shot was to generate or edit a reference frame, check it, then hand it to an image-to-video model, and gpt-image-1-mini made the draft half of that loop cost half a cent an image. The retirement is unremarkable as model churn goes, and the six-month-plus notice is comfortably inside what OpenAI promises for generally available models. What makes it worth recording is where the breakage sits. The model string is the easy part. The parts that cost real work are an endpoint that quietly drops off the replacement's support table and a pricing unit that changes from images to tokens, and neither of those appears in the deprecation announcement. You only see them by opening the model page for the thing you are migrating to and reading the endpoint and pricing tables next to the ones for the model you are leaving. That is the habit worth keeping: when a provider names a replacement, treat the name as a starting point for a diff, not as a promise that your integration shape survives. A replacement model is a replacement for the model, not for the plumbing you built around it.