Is OpenAI (ChatGPT) down?

Understanding OpenAI and ChatGPT availability

This page covers two related but separate access paths: the ChatGPT interface and the OpenAI API at api.openai.com. An Up result means the monitored endpoint responded when it was checked. It does not confirm that every OpenAI product, model, account, region, or individual request is operating normally.

A wider service issue is more plausible when unrelated requests, users, or applications fail together, latency increases, or repeated 502 Bad Gateway, 503 Service Unavailable, or timeout responses occur across different environments. A single 429 Too Many Requests response usually points first to request frequency, concurrency, token usage, or account-level quota rather than a complete outage.

ChatGPT can work while an API integration fails because authentication, limits, endpoints, request parameters, and network paths differ. The reverse can also occur. Compare the timing and scope of recurring symptoms with the official OpenAI status page, but treat that page and any endpoint check as evidence about availability—not proof that every account or operation is healthy.

Troubleshooting OpenAI or ChatGPT errors

1. Check whether a wider service issue is plausible

Compare the failure pattern across unrelated requests, users, applications, or networks. Repeated 502, 503, or timeout responses across more than one environment provide stronger evidence of a service-side problem than a single failed request. Check the official OpenAI status page for incident information.

For an application, use controlled retries with increasing delays rather than repeatedly sending requests at high frequency. Record the response status, time, endpoint, model, response body, and any request identifier returned by the service. Do not retry state-changing or costly operations without confirming that the original request did not complete.

2. Separate a local or network problem

If the service check reports Up but your request fails, test from another authorized network, such as a mobile connection, and—where appropriate—retry without a VPN or proxy. Check whether the affected device can resolve the relevant hostname, establish outbound HTTPS connections, and reach api.openai.com.

A timeout or connection refusal limited to one device, office network, firewall, proxy, DNS path, or routing path points more strongly to local connectivity than to a global OpenAI outage. Do not change firewall, proxy, or network policy without authorization.

3. Check account, quota, and request causes

• 429 Too Many Requests: review request frequency, concurrent requests, token usage, and the API account's applicable limits or quota.
• Authentication or permission errors: confirm the intended credentials, project or organization context, endpoint, and model access. Rotate exposed credentials through the approved account process; never print API keys in logs.
• One model or operation fails: inspect the model name, input size, parameters, endpoint, and application logs before treating the symptom as platform-wide.
• Malformed or rejected requests: compare the request method, headers, body, content type, and parameter values with the applicable OpenAI API documentation.

4. If the ChatGPT browser interface is affected

If ChatGPT fails while the API works, try a private browsing window, temporarily disable extensions that modify pages or requests, clear the site's stored data, and test another browser. If the issue is limited to one browser profile or device, a session, extension, stored-data, or local network problem is more likely than a service-wide outage.

5. If an API integration is affected

With appropriate authorization and without logging secrets, test a minimal request outside the application's normal code path. If it succeeds, inspect the application's timeout settings, proxy configuration, retry behavior, concurrency, response parsing, and credential-loading path. If it fails consistently from independent environments, preserve the endpoint, model, timestamp, status, response body, and request identifier for comparison with the official status page or escalation to the vendor.