Is OpenAI down?
Right now
Partly. OpenAI is degraded.
Delayed Costs data in the Compliance API.
Read from status.openai.com · every 60s
For agents
proceed
Proceed, expect slow or flaky responses.
GET upbutler.com/status/openai/status.agent.json
MCP: upstream_status {"service":"openai"}
Chat Completions, Responses, embeddings and ChatGPT. An outage fails every feature that calls a GPT model, usually as 5xx or timeouts on api.openai.com.
What OpenAI reports
Degraded· 1 open incident
Taken as is from OpenAI's own status page API. Incident titles and component names below are theirs.
Open status.openai.comUpButler's independent probe
API edge is answeringHTTP 401 · 209 ms
One request without credentials to api.openai.com every two minutes. An answer (the expected "not authenticated") shows the edge, TLS and routing work. It does not prove that authenticated calls succeed.
Checked · up since 2026-10-10 01:23 UTC
Open at OpenAI
- Delayed Costs data in the Compliance APImonitoring · impact minor
Since
We have implemented a mitigation, and delayed Costs data is now processing through the Compliance API. Some customers may continue to see incomplete Costs data while the backlog catches up. We expect this catch-up process to continue through Sunday, October 11. ChatGPT usage and API Platform requests remain unaffected. We are monitoring recovery.
Text written by OpenAI, shown unchanged.
OpenAI components
1 of 25 not operational
- Chat CompletionsOperational
- ResponsesOperational
- AgentsOperational
- DecisionsOperational
- Fine-tuningOperational
- EmbeddingsOperational
- ImagesOperational
- BatchOperational
- AudioOperational
- ModerationsOperational
- RealtimeOperational
- FilesOperational
- LoginOperational
- ConversationsOperational
- LoginOperational
- ChatGPT WorkOperational
- Codex in ChatGPT DesktopOperational
- Compliance APIDegraded
- SearchOperational
- File uploadsOperational
- Voice modeOperational
- GPTsOperational
- Image GenerationOperational
- Deep ResearchOperational
- AgentOperational
OpenAI incidents, last 90 days
24 finished incidents seen since UpButler started tracking on 2026-10-10, 4 of them rated major or critical by OpenAI. A typical one lasted 2 h 4 min; the longest 45 h 24 min.
Tell your outage from OpenAI's
Declare that your project depends on OpenAI. When one of your monitors fails while OpenAI has an incident that began around the same time, UpButler tags your incident "likely upstream" with their incident linked, says so in the alert and the AI report, and tells your on-call agent not to change code.
npx upbutler init finds the dependency from your env var names and packages. The catalog, dependencies and upstream-aware incidents are on every plan, including Free.
# upbutler.yaml
dependencies:
- openaicurl -s https://upbutler.com/status/openai/status.agent.json{
"schema": "upbutler.upstream-status/v1",
"service": "openai",
"verdict": "proceed",
"status": "degraded",
"reason": "Delayed Costs data in the Compliance API",
"retryAfterSec": null,
"incident": {
"title": "Delayed Costs data in the Compliance API",
"url": "https://status.openai.com/incidents/01M4GV1M2A3K38T73JCVCD9G0R"
},
"probe": {
"state": "up",
"statusCode": 401
}
}Questions
- Is OpenAI down right now?
- Partly. As of 2026-10-10 05:12 UTC, OpenAI is degraded, with an open incident: "Delayed Costs data in the Compliance API". This page reads the status again every 60 seconds.
- Where does this status come from?
- From OpenAI's own status page API at https://status.openai.com. UpButler reads it once for everyone (not once per visitor) every 60 seconds with conditional requests, and shows it unchanged: incident titles and component names are OpenAI's words.
- What does UpButler's own OpenAI check prove?
- Every two minutes UpButler sends one request without credentials to the OpenAI API edge (api.openai.com). Any answer from the edge, usually HTTP 401, shows that DNS, TLS and routing work. It does not prove that authenticated requests succeed, so it can say "reachable" while OpenAI reports degraded service. It is useful the other way round: when the probe fails before their status page changes.
- How can my agent or app check OpenAI status?
- Fetch https://upbutler.com/status/openai/status.agent.json: a small JSON document with a verdict (proceed, retry, fallback or pause) and retryAfterSec. No key is needed. The same answer is an MCP tool, upstream_status {"service":"openai"}, on https://upbutler.com/mcp/public.
- How do I know whether an outage is mine or OpenAI's?
- Add "openai" to the dependencies of your UpButler workspace. When one of your monitors fails while OpenAI has an incident that began around the same time, your incident is tagged "likely upstream" with their incident linked, and your on-call agent is told not to change code.
More in ai models and tooling
UpButler is not affiliated with OpenAI. Status and incident text belong to OpenAI and are read from their public status source; the probe result is UpButler's own measurement.