OpenAI Status: Is ChatGPT Down Right Now?

Check OpenAI status in 30 seconds. ChatGPT is either down for everyone or broken only for you, and these three links tell you which. Open them in this order:

1. Open the official OpenAI status page at status.openai.com. If ChatGPT is down for everyone, a red or amber incident banner appears here.
2. Read the OpenAI incident history. It shows whether this has been going on all morning.
3. Compare user reports on Downdetector’s OpenAI outage page or StatusGator’s ChatGPT status tracker. These show user-reported spikes, which usually move 5 to 15 minutes ahead of an official page.

If all three look clean, the fault is almost certainly on your side, and the five checks below will find it in about two minutes. If Astra is missing rather than broken, that is a rollout question: see how to access ChatGPT Astra, or our GPT-6 Astra explainer.

How do I check OpenAI status in 30 seconds?

Load status.openai.com and read the banner at the top. Green means all systems operational. Any amber or red entry names the affected component and carries a timestamped update log. That page answers “is ChatGPT down” in most cases.

It splits OpenAI’s services into separate components, which matters more than people expect. ChatGPT can be operational while the API is degraded, or the reverse. If your app is failing, read the API row instead of the overall colour. Subscribing to updates there is worth 30 seconds.

Should I trust status.openai.com or Downdetector?

Use both. The official OpenAI status page is authoritative but slow, because someone must confirm an incident before publishing it. Downdetector and StatusGator are fast but unverified, since they aggregate user complaints. Fast tells you something is wrong; authoritative tells you what.

Third-party trackers count problem reports against a baseline, so a sudden vertical spike is a strong signal before OpenAI acknowledges anything. The weakness is that spikes also follow a viral post, a regional ISP fault, or an unrelated app that embeds ChatGPT. StatusGator and IsDown add a second angle by aggregating official status pages across many services at once.

Is it OpenAI or is it me? Five checks that tell you which

If the official page is green, run these checks in order. Most “ChatGPT not working” reports resolve at step one or two.

Step 1: Try a different network

Switch from Wi-Fi to mobile data, or back. If ChatGPT loads on one and not the other, your network is the problem.

Step 2: Open a private browsing window

This disables most extensions and ignores cached files. If ChatGPT works there, an extension or stale cache is the culprit.

Step 3: Turn off your VPN

Shared exit addresses get flagged for suspicious traffic, so OpenAI’s edge protection may serve an access-denied page while the service is healthy. ChatGPT then looks broken only for you.

Step 4: Log out and back in

Session tokens expire. A stale one produces login loops, blank conversation lists and the vanishing chat history people mistake for data loss.

Step 5: Check your account state

A message cap, an expired card or a plan change looks identical to an outage from the inside. Business and Enterprise workspaces add a layer, since admins must enable newer models first.

What do the common ChatGPT error messages actually mean?

Each error points at a different layer. “Something went wrong” is generic and usually transient. A 502 or 503 means OpenAI’s servers are struggling. A 429 means you hit a limit. Access-denied usually means network or VPN filtering. Reading the code stops you fixing the wrong thing.

What happened in the 3 September 2026 ChatGPT, Claude and Grok outage?

Three major AI chatbots failed within hours of each other. ChatGPT was affected from roughly 7:43am PT to 8:17am PT, with Codex hit too. Claude was down 3 hours 6 minutes, restored at 16:16 UTC. Grok’s investigation opened at 6:30am PT. The Register reported the overlap the same day.

SpaceX attributed the disruption to an outage at its Memphis compute centre, which affected Grok and what it called “impacted compute partners”, including Anthropic, which shares infrastructure there. Cloudflare, often blamed when large sites fail together, denied involvement. AWS, Google Cloud and Azure status pages showed no relevant issues.

Note the asymmetry: OpenAI’s outage was short and has not been publicly tied to the Memphis incident. Simultaneous does not mean identical cause.

Why do ChatGPT, Claude and Grok go down at the same time?

Because the industry rents from a small number of suppliers. Model providers share data centres, GPU capacity, network transit, DNS and edge protection. When one layer fails, several unrelated-looking chatbots fail together, even though rivals built them.

The 3 September event shows this cleanly. Anthropic ran capacity on infrastructure operated by SpaceX, and that relationship was invisible to users until it broke. So “use a different AI as backup” is weaker redundancy than it looks, because two chatbots from rival companies can sit on the same floor.

The same dependency question applies to any hosted platform your team relies on, which is why our guide to LMS architecture explained is worth reading before you assume a vendor’s uptime is entirely its own. Ask who actually runs the compute, the storage and the content delivery layer. The related question of LMS scalability and how many users a platform can really handle comes from the same place: shared capacity behaves well until everyone needs it at once.

Did the GPT-6 Astra launch strain OpenAI’s capacity?

This is an open question, not a confirmed cause. Astra was announced on 3 September 2026, the same day as the outages, and a major launch does drive a demand surge. But OpenAI has not attributed its incident to Astra traffic, and the only published root cause that day points at SpaceX.

Treat any link between the two as speculation until OpenAI says otherwise. Correlation on a single day is thin evidence, and Astra’s rollout was deliberately staged, starting with a limited “Daybreak Access” group. If capacity pressure does contribute to a future incident, expect degraded performance and rate limiting rather than a hard outage.

What should I do while ChatGPT is down?

Stop refreshing and switch modes. Refreshing a failing page adds load without helping. Confirm the incident, note the time, and move to the part of your task that does not need a model. Most outages this year cleared within an hour.

• Copy your last prompt somewhere safe before closing the tab. Unsent drafts do not survive a session reset.
• Subscribe to the incident on status.openai.com so you know when it clears.
• Try a second provider, after checking its own status page.
• On the API, back off exponentially rather than retrying in a tight loop, which delays recovery.
• Do the review and verification work that needs judgment, not a model.

How do I build a workflow that survives AI downtime?

Assume the model will be unavailable occasionally and design for it: a second provider credentialed in advance, retries with backoff, prompts stored outside the chat window, and no step that silently fails when the API returns 503.

Teams that shrug off an outage do three things. They keep prompts and context in a document, not only in chat history. They treat switching provider as a config change, not a project. And they know which workflows are genuinely time-critical, because most are not. This matters more as models take on the longer autonomous jobs covered in our guide to GPT-6 Astra use cases, where an outage mid-run costs more than a failed message. Checkpoint long runs so they resume instead of restarting.

Picking the standby matters as much as having one. Choose it before you need it, test one real task on it, and keep the account signed in, which is easier if you start from a shortlist of free AI tools that cost nothing to keep warm.

OpenAI outage log (updated)

We keep this log current and add incidents as providers confirm them.

3 September 2026: ChatGPT, Codex, Claude and Grok. ChatGPT down from about 7:43am PT, resolved about 8:17am PT. Claude down 3 hours 6 minutes, restored 16:16 UTC. Grok investigation opened 6:30am PT. Root cause per SpaceX: an outage at its Memphis compute centre affecting Grok and impacted compute partners including Anthropic. Cloudflare denied involvement. Cloud status pages showed no relevant issues.

Last reviewed 5 September 2026. This log is maintained and updated as new incidents are confirmed.

Conclusion

Make the next outage cheap. Bookmark the OpenAI status page, subscribe to its notifications, and put the Anthropic Claude status page and the xAI Grok status page beside it, so you can separate a single-provider fault from an industry-wide one in seconds.

Then pick one workflow you would miss for an hour and wire up a second provider for it today, while nothing is broken. That is a 20-minute job on a calm afternoon and an impossible one at 7:43am during an incident.