When the image API returns 5xx
separate outage, policy, and quota
A failed ChatGPT or Images API render is not always your prompt. Read the official status page for elevated errors, then check policy filters and plan caps. QCode image model ids show 0 calls in 30 days, so this page does not claim image generation is on sale here — only the triage order.
Updated 2026-09-11
Pin these four facts
Server-side elevated errors
When the status page says elevated errors affecting image generation, some requests are expected to fail. That is not a typo in your key.
Policy or bad parameters
A blocked prompt, a rejected reference image, or an unsupported size is usually not 5xx. Do not treat a policy refuse as “the API is down”.
App image caps or API credits
ChatGPT plans have image-count windows; the API is a different ledger. Exhausted caps and a status incident are not the same outage.
Image ids at 0 for 30 days
gpt-image-* rows on QCode are 0. Category is troubleshooting. We do not say an image model is already available here.
How to handle this class of image failure
Elevated errors on image generation keep coming back; this is not a one-day news item. Order of operations: open status.openai.com and see whether ChatGPT, the Images API, or both are named; replay one tiny prompt and read 5xx vs 429 vs 400; only then change endpoints or task type. QCode can keep a text path alive. It cannot honestly claim a specific image id is on sale.
What 2026-09-08 wrote
On 2026-09-08 MacRumors quoted the OpenAI status page: We're investigating elevated errors affecting image generation in ChatGPT and the Images API, plus Some requests may fail. File uploads were slow in parallel. This page does not say the incident is still live — the timeline keeps the date, the evergreen section keeps the triage.
Timeline
2026-07-21 a third-party status mirror logged Elevated errors for API image generation for a few hours. Image 5xx repeats; it is not a single headline.
2026-09-08 the status page investigated elevated image errors in ChatGPT and the Images API. MacRumors quoted the investigating line and Some requests may fail.
On 2026-09-11 do not copy a history row into “still down right now”. Trust the live status sentence. Evergreen copy stays a three-way split.
Confirmed versus easy to over-read
Sentences you can point at
On 2026-09-08 the status page was quoted verbatim: investigating elevated errors affecting image generation in ChatGPT and the Images API. Some requests may fail sat in the same update. Image components and text chat are listed separately.
What the status page did not say
Some threads treat one image 5xx as “all of GPT is down” or “your API key is banned”. The status sentence names image generation. Unconfirmed: how many days any one incident lasts, or whether it hits only the free tier. Use the live status sentence, not a social screenshot. 2026-09-11
Two paths
Wait on the official image component
Fits while the status page still says investigating or monitoring. Replay one tiny prompt. Do not pile retries into 429.
Keep a text path alive
Fits when you need to keep working more than you need that one image. Same API key, switch model to gpt-5.6-sol or claude-sonnet-5, confirm HTTP 200. That is not “QCode will draw the same picture”.
Three steps when you are stuck
First: open status.openai.com and copy the image-related sentence; do not close the case from a red bubble in chat. Second: one tiny prompt, once. 5xx waits on status; 429 means cut concurrency; 400 means change prompt or size. Third: if the job can become text, point Codex CLI or your HTTP client at a text model that still returns 200.
What to do on QCode
Image ids show 0 calls in 30 days — no image-sale claim. Keep text working: copy the dashboard key after signup. Claude Code: ANTHROPIC_BASE_URL = https://api.qcode.cc and ANTHROPIC_AUTH_TOKEN set to that key. Codex CLI: OpenAI-compatible base URL on the same host, model gpt-5.6-sol first (official $4 / $20, promo through at least 2026-11-21). Next step: one non-streaming tiny request, HTTP 200, matching model id.
FAQ
Does a failed image always mean an outage?
No. Check the status page for image elevated errors, then the HTTP code. 5xx vs 429 vs 400. A policy refuse is not an outage.
If ChatGPT’s website fails, is the API down too?
Not necessarily. The status page lists ChatGPT and the Images API as separate components. When the website fails, read the API line.
Is QCode an image fallback?
We do not claim that. gpt-image-* usage is 0. “Already generates images here” would be false.
Do retries help?
Often no. Tight 5xx retries become 429. Wait until the status page leaves investigating, then send the same tiny prompt once.
Is this the same as the GPT-6 launch outage?
Not always. Image elevated errors showed up in 2026-07 and 2026-09. Treat them as a recurring class, not one GPT-6 incident.
How do I probe a text model instead?
Stop using ChatGPT login. Point the OpenAI-compatible base URL at https://api.qcode.cc , use the dashboard key, model gpt-5.6-sol or claude-sonnet-5.
Sources
2026-09-08 investigating sentence: MacRumors quoting We're investigating elevated errors affecting image generation in ChatGPT and the Images API. History shape: status.openai.com/history. QCode image-id zeros are internal CRS, not an external number. Fetched 2026-09-11.
On image 5xx, keep the text path
One key for gpt-5.6-sol and claude-sonnet-5. No image-id claim. Plans from ¥60/month.
Related
OpenAI API outage checklist
5xx / 429 / quota, status page first.
Why GPT-6 Astra burns quota
Do not read an empty credit bucket as an image outage.
When ChatGPT Pro is unavailable
Subscriptions and APIs are different products; a failed image is not a reason to buy a plan.
Triage notes, not an OpenAI status statement, and not a claim that QCode lists image models. Status page and /models win.