Blog / Troubleshooting

Is ChatGPT Down or Is It Your Network? A 3-Step Check

2026-10-17cdnping editors7 min read

When ChatGPT won't load, the first question is usually "Is ChatGPT down, or is it my network?" Check in this order. First, look at the official status page and other users' reports to see whether it's a server-side outage. Second, look at an independent API probe to see whether OpenAI's entry point is reachable. If both look normal, the problem is most likely your own exit IP, DNS, or browser environment. Each of the three steps maps to a page on cdnping, and the whole check usually takes a few minutes.

#How often does ChatGPT have incidents?

According to the OpenAI official status page history that cdnping syncs, OpenAI published 25 incidents in the 30 days from 2026-09-17 to 2026-10-17:

Here are the 5 incidents from 2026-10-13 to 10-15:

• 10-15, "Widespread feature degradation", major, about 146 minutes.
• 10-14, degraded performance in Codex and ChatGPT Work, minor, about 167 minutes.
• 10-14, limited availability of some image models, minor, about 40 minutes.
• 10-14, delayed cost data in the Compliance API, minor, about 42 hours; this affected only the enterprise compliance interface.
• 10-13, degraded performance in Codex and Work mode, minor, about 506 minutes.

Most incidents are minor, and many affect only one feature or one class of models, which is not the same as ChatGPT being unreachable. There were 4 major incidents in 30 days. So when ChatGPT won't load, a server-side outage is one possibility, but verify it first rather than assuming it.

These incidents come from syncing the official status page history. cdnping's own probes only started collecting on 2026-10-17, so there are no long-term statistics yet.

#Step 1: What do the official status and other users say?

Open /en/down/chatgpt. This page puts three kinds of signals side by side:

1. The OpenAI official status page, which cdnping syncs every 15 minutes.
2. API probes, which hit OpenAI's official endpoint from the Cloudflare edge every 5 minutes.
3. One-click user reports from other people: can't open, slow, "got dumber", rate limited, quota reset, logged out.

Read the combination like this:

Official announcements sometimes lag behind what users actually experience, so pay extra attention when the status is normal but reports are spiking. If you're affected too, click the report button; it helps the next person judge faster.

The official status page itself is at status.openai.com.

#Step 2: Is OpenAI's API endpoint reachable?

Open /en/status/openai. This is cdnping's connectivity probe from the Cloudflare edge to api.openai.com, with 30 days of history (the history builds up as data is collected).

The API probe tests api.openai.com, while the ChatGPT web app uses chatgpt.com. They are different entry points, so read this step together with step 1:

• API probe fails and the official status shows an incident: you can be fairly sure it's a server-side problem.
• API probe is normal but the web app won't load: the problem is more likely in the web app, your account state, or your network path to chatgpt.com.
• API probe latency suddenly rises: the service is fluctuating, and the web app may feel slow to load or cut off answers midway.

For reference, in cdnping's first batch of edge probes on 2026-10-17 (the Cloudflare Denver node, a window of about 85 minutes), every probe of the OpenAI API succeeded, with a median round trip of 96 ms and a P95 of 194 ms. The window is short; it only shows the endpoint was reachable at collection time and says nothing about long-term behavior.

The probe nodes are overseas. They test whether OpenAI's entry point itself is healthy, not whether it's healthy from your network. You need step 3 to check that part yourself.

#Step 3: How do I check my own exit IP and DNS?

If the first two steps look normal, check your own side. Open /en/ip and go through the checklist below.

#Checklist

• Exit IP location and type: the page locates your IP by voting across 11 geolocation databases and gives a confidence level. It also labels the IP type (datacenter, residential, mobile, Tor). OpenAI is available only in certain countries and regions, and access is refused when your exit IP geolocates outside the supported area.
• IP reputation: the page aggregates 14 reputation sources and publishes the weight of each hit. If your exit IP appears on many blocklists, you may hit CAPTCHAs or block pages often.
• Multi-exit detection: the page reads the Cloudflare trace from sites such as ChatGPT, Claude, and Grok to see which exit you actually use for each. If ChatGPT's exit differs from the others, your network is sending ChatGPT traffic down a different path.
• Mainland China split-routing detection: determines whether you're in "mainland split routing / global proxy / direct domestic connection" mode, so you can confirm how your traffic actually flows. (In mainland China, many proxy setups route domestic and foreign traffic differently.)
• DNS leak and WebRTC leak: shows who actually resolves your DNS requests. A misbehaving resolver can return wrong addresses, which shows up as pages not loading or certificate errors. For background, see DNS leaks and WebRTC leaks.

#Generate a result image to ask for help

After /en/ip finishes its checks, you can generate a watermarked result image, and you can choose to hide your IP address before generating it. When you ask a coworker or a community for help, this image is clearer than a text description, and hiding the IP keeps your exit address from becoming public. Browser fingerprinting is computed locally and is not uploaded.

#Command-line checks

If you can't easily open a web page, run these commands in a terminal:

# 1. Compare DNS results from the system default resolver and a public resolver
nslookup chatgpt.com
nslookup chatgpt.com 1.1.1.1

# 2. See the exit used when reaching chatgpt.com (the ip and loc fields in the response)
curl -s https://chatgpt.com/cdn-cgi/trace

# 3. Check whether the API endpoint is reachable, and the time spent in each phase
#    Without a key, a 401 is normal and means the path is working
curl -s -o /dev/null \
  -w "HTTP %{http_code}  DNS %{time_namelookup}s  connect %{time_connect}s  TLS %{time_appconnect}s  total %{time_total}s\n" \
  https://api.openai.com/v1/models

How to read the results:

• If the two resolvers return very different addresses, start by checking the DNS settings on your machine or router.
• In cdn-cgi/trace, ip is the exit address the other side sees and loc is the country or region code Cloudflare assigns to it. Go by these two fields.
• If the third command stalls at the DNS or connect phase, it's a network-layer problem. If you get a 401, the network is fine and the problem is in your account, browser, or the web app.

#What do common errors usually mean?

About regional restrictions: OpenAI's service has a defined regional scope, and "Unsupported country / region" is a policy restriction. This article only covers how to tell which side the problem is on, and does not discuss how to bypass such restrictions. Please follow OpenAI's terms of service and local laws.

#Next time it won't load, where should I look first?

• Bookmark /en/down/chatgpt and check it first when ChatGPT won't open.
• If you mainly use OpenAI through the API or Codex, /en/status shows API status for all connected providers, so you can tell whether one provider is down or there's a wider network problem. AGI Plan is a commercial partner of cdnping, and AGI Plan's public uptime is also mirrored at /en/status/agiplan.
• On the home page, /en/, run an online speed test to see the first-connection and round-trip times from your current network to OpenAI and other AI APIs.
• To keep up with service stability over time, subscribe to the weekly stability report, generated automatically every Monday. The probing method is described at /en/methodology.