- Audio Operational
- Batch Operational
- Chat Completions Operational
- Embeddings Operational
- Files Operational
- Fine-tuning Operational
- Images Operational
- Login Operational
- Moderations Operational
- Realtime Operational
- Responses Operational
- Sora Operational
Covers the official API components currently exposed by OpenAI. It does not test an authenticated customer request or model generation.
OpenAI API Status for Developers
The live card above summarizes API-related components currently exposed in OpenAI’s public status feed. API availability can differ from consumer ChatGPT availability, and a healthy public component does not verify your account, API key, model, region, or usage tier. The API is a distinct component that can fail independently of the consumer-facing chatgpt.com product — we surface API-specific incidents below.
Is the API Actually Down, or Is It My Code?
Run these quick checks before debugging your own code:
- Hit the models endpoint:
curl https://api.openai.com/v1/models -H "Authorization: Bearer $OPENAI_API_KEY"— a 200 response means auth and basic connectivity are fine. - Check your key on the OpenAI dashboard — revoked, expired, or rate-limited keys return 401 or 429.
- Check your billing — a zero-balance account returns a specific error, not a generic 500.
- Check your rate limits — burst traffic triggers 429 even on healthy infrastructure.
- Try a different model — a model-specific rollout can produce 400 or 404 while others work.
Common API Error Codes
- 401 Unauthorized — invalid or revoked API key.
- 429 Too Many Requests — rate limit exceeded. Back off with exponential delay.
- 500 Internal Server Error — OpenAI infrastructure issue. Retry with backoff; if it persists check the live status above.
- 502 / 503 / 504 — upstream gateway issue, usually transient. Retry after 30–90 seconds.
- 400 Bad Request — your payload is malformed; not an outage.
Incidents That Affect Only the API
API incidents can affect developers even when the consumer ChatGPT website remains available. Check the component-specific incidents below and confirm affected endpoints against OpenAI's official status updates.
Response Latency and Your Code
API latency varies by model, request size, output length, account tier, region and current load. Compare your authenticated request timing with official incident updates and the official rate-limit guide; our homepage chart measures reachability to chatgpt.com, not API completion latency.
Frequently Asked Questions
Is the OpenAI API down right now?
Check the live component card at the top of this page. It summarizes API-related components from OpenAI's public status feed. Confirm any active incident on OpenAI Status, then compare the affected components with your own authenticated request.
Why am I getting 429 errors when traffic is low?
OpenAI enforces both per-minute and per-day rate limits, and they vary by model and account tier. A quiet-looking spike in concurrent requests can still trigger the per-minute limit.
Does the OpenAI API have a public SLA?
Availability commitments, if any, depend on the customer's contract or applicable service terms. Check your agreement and the current OpenAI Services Agreement rather than relying on a universal public SLA.
Sources and scope
The live card reads the public OpenAI component feed. It does not send an authenticated request through your API account, model, region or usage tier.