Image API changes in October 2026: what to do and when
October 2026 image API changes: gpt-image-1 shuts down 2026-10-23, three more OpenAI ids on 2026-12-01, GPT Image 2.5 ships, and Sume retires Image 1.0.

The image API change with a deadline is OpenAI's shutdown of gpt-image-1 on 2026-10-23, 18 days from today. Three more ids, gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest, shut down on 2026-12-01. OpenAI's deprecations page (read 2026-10-05) names gpt-image-2.5-sunburst and gpt-image-2.5-flare as the replacements for all of them.
Alongside the deadlines, there is a new model family to choose from, a note from xAI on image editing, and a Sume retirement. This post lists each change, the date and the action.
The changes in date order
Dates are from the vendors' pages, read today, except the Sume row, which is from Sume's own docs.
| Date | Change | Action |
|---|---|---|
| 2026-09-08 | GPT Image 2.5 released as Flare and Sunburst (OpenAI changelog) | Test both ids |
| 2026-10-23 | gpt-image-1 shuts down (OpenAI deprecations) | Replace id, check parameters |
| 2026-12-01 | gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest shut down | Replace in the same pass |
| Soon | Sume retires Image 1.0; its public URLs stay aliases for Auto | Move to an explicit model id |
What changes in your requests
Moving to GPT Image 2.5 on Sume changes more than the id. quality takes auto, low, medium, high, xhigh and max, and defaults to high. Custom sizes follow the multiple-of-16 rule with a 3840 edge cap. Parameters a model does not advertise return 400 unsupported_parameter, and output is a Sume-hosted URL in data[].url.
A call that runs longer than 30 seconds returns 202 and a job envelope, so any client that expects an image in the first response needs a polling path. The Jobs and results page describes it.
Prices to plan with
GPT Image 2.5 at 1024x1024 on Sume quotes by tier.
| Quality | Sume price |
|---|---|
| low | $0.0074 |
| medium | $0.0165 |
| high | $0.0659 |
| xhigh | $0.1171 |
| max | $0.2635 |
Other image news
xAI's release notes (read 2026-10-05) say image editing on Grok Imagine Image 2.0 now accepts up to 5 source images per request, up from 3. On Sume the Grok Imagine row quotes a flat $0.0250 per image and returns one image per call. Check the live catalog before you rely on a reference count.
For reference, Nano Banana 2 on Sume quotes $0.0750 to $0.2000 by tier and Nano Banana Pro $0.1875 to $0.3750, per the catalog today.
On Sume you can also send model: "sume/auto" and let Sume pick the family. Its Image API page says auto routing continues to use Flare and the response does not reveal which family ran, so use an explicit id wherever you need to know.
A checklist
Work through these in order.
- Search every repo, fixture and config for the four retiring ids.
- Replace them with
openai/gpt-image-2.5or the Sunburst id and setqualityandimage_sizeexplicitly. - Run one call per code path and compare
usage.costwith the table. - Handle
202responses. - Put the id in one config value so the next change is a one-line edit.
What not to change
Do not rewrite your prompts at the same time as the id. If you change the model and the prompt together and the output looks worse, you cannot tell which change caused it. Move the id first, compare outputs at the same prompt and tier, and then tune.
Do not leave the December ids for December. They retire with the same replacements as gpt-image-1, so the only extra cost of moving them now is a scan that you are already running. Doing it in one pass leaves a single clean date to remember, and the cutover calendar in your own tracker shrinks to a single line.
Do not trust a green test run on a mocked client. A fixture that returns a canned image will pass on October 24 exactly as it does today. Run at least one real call per code path before the date, and keep it in your release checklist.
Sources
Related posts
More in Developers
- Retry an image edit after a timeout: same key, same payload
After a client timeout on an image edit, resend the same payload with the same Idempotency-Key. If you change the prompt, change the key.
- image-generations_create is now generate_image: fix old MCP prompts
The Sume MCP tool image-generations_create was retired for generate_image. Update agent prompts and allow-lists, and check with tools_list. Dated facts.
- MCP: read image-models_get, then dry-run generate_image first
An agent checklist for hosted Sume MCP: read the model with image-models_get, dry-run generate_image with a spend cap, then submit with a new idempotency key.
- A 4K image request returns 202: poll the job and fetch the result
POST /v1/images waits 30 seconds. Slow 4K or high-quality calls return 202 with a job envelope. Python that handles both and polls to completion.
Written by Sume