Is ChatGPT Down? Inside the September 2026 Outages and Codex 404 Errors
If ChatGPT has been throwing 404 errors, getting stuck on "Reconnecting," or failing to load chatgpt.com entirely over the past few days, you're not alone. Early September 2026 has brought a fresh cluster of disruptions, some acknowledged on OpenAI's own status page and some only visible through independent trackers and user reports. Here's what's actually confirmed, what's still murky, and what to do while you wait it out.
A direct hit to chatgpt.com's backend-api path returning a plain 404 during the incident window, screenshotted September 3, 2026.
Partly - and It Depends When You Check
As of early September 2026, ChatGPT hasn't had one single, hours-long global blackout the way it has during some past incidents. Instead, OpenAI's own status page shows a run of shorter, more scattered problems: an "Elevated latency in the Responses API" incident and a separate "Elevated errors in ChatGPT conversations for Free and Go plans" incident on September 1, an "Elevated errors creating new accounts" issue on September 2, and a "ChatGPT Work Mode High Error Rates" incident that OpenAI marked resolved in the early hours of September 3.
The same 404 also showed up on chatgpt.com's plain homepage URL, including links loaded straight from a paid Google Ads click, ruling out a one-off broken deep link.
At the same time, independent outage-tracking sites have continued to show elevated user complaints later in the day on September 3, including reports of chatgpt.com failing to load and Codex requests returning 404 errors. That gap between "officially resolved" and "still being reported" is common with short, partial, or regional incidents - the status page reflects OpenAI's own monitoring thresholds, while trackers reflect whatever a slice of users happen to report, which doesn't always line up in real time.
The 404s, and the Endless "Reconnecting"
Inside the chat window itself: a "Reconnecting 5/5" loop paired with a raw 404 error on the Codex responses endpoint, cf-ray ID and all.
Two specific symptoms have been widely reported around this incident window. The first is a "Reconnecting" loop in the ChatGPT and Codex interface, sometimes followed by an error referencing an "unexpected status 404" on a backend API route - a sign that a specific request path, not the whole site, is failing to respond. The second is chatgpt.com itself returning a plain 404 "This page can't be found" error, which points to a routing or DNS-level problem rather than an issue inside a logged-in session.
These two symptoms usually have different causes. A 404 on an internal API path (like a Codex "responses" endpoint) typically means a backend service behind that specific feature is degraded or was briefly redeployed - consistent with the kind of "elevated error rate" incidents OpenAI has logged repeatedly through 2026. A 404 on the bare domain itself is more consistent with a CDN, DNS, or edge-routing issue, which can affect some users or regions while leaving others unaffected, and often clears on its own within minutes once traffic is rerouted.
Two incidents, same day
OpenAI logs elevated latency in the Responses API and, separately, elevated conversation errors for Free and Go plan users. Both are marked fully recovered the same day.
Account creation errors
New account creation sees an elevated error rate for several hours before OpenAI marks the issue resolved that evening.
Work Mode errors, then user reports continue
A "ChatGPT Work Mode High Error Rates" incident is marked fully recovered just after midnight. Independent trackers and user reports of chatgpt.com and Codex failures continue later in the day, without a matching new incident yet posted to the official status page.
Part of a Longer 2026 Pattern
This isn't an isolated week. OpenAI's own status history shows roughly 19 separate incidents logged in August 2026 alone, ranging from image-generation errors and file-upload failures to login problems and Codex authentication issues. Every major AI provider has had a rough reliability year in 2026 as usage has scaled faster than infrastructure teams can always keep up with - it's an industry-wide pattern, not something unique to any one company.
What makes days like this confusing for users is the mismatch between official acknowledgment and lived experience: a status page can mark an incident "resolved" while a meaningful number of people are still seeing errors, either because the fix hasn't fully propagated, a new but related issue has started, or the disruption is regional enough to fall under OpenAI's own reporting threshold.
What to Do While It's Down
- Check status.openai.com for an active, acknowledged incident before assuming it's your connection
- Try a different browser, device, or network (like a mobile hotspot) to rule out a local or ISP-side issue
- A 404 on a specific feature (like Codex) doesn't always mean the rest of ChatGPT is affected - try the plain chat interface separately
- Avoid repeatedly resubmitting the same request - that adds load to a system that's already struggling
- For time-sensitive work, keep a fallback plan rather than waiting indefinitely on one tool
- Cross-check an official status page against a live outage tracker rather than trusting either alone
- If the issue is isolated to your account (rather than a broader incident), logging out and back in occasionally clears it
Frequently Asked Questions
Is ChatGPT down right now?
It depends on the moment you check. OpenAI's own status page has logged several separate incidents across September 1-3, 2026, each marked resolved within hours, while independent monitors have continued to show elevated user reports afterward. Check status.openai.com directly and compare it against a live tracker before assuming it's just your connection.
Why does ChatGPT show a 404 or "page can't be found" error?
A 404 on chatgpt.com usually means the request never reached a working backend - either a specific API route is temporarily unavailable during an incident, or a DNS/CDN routing issue is misdirecting the request. It's rarely something wrong with your account, and it typically clears once OpenAI resolves the underlying incident.
Why is Codex or ChatGPT stuck on "Reconnecting"?
The Reconnecting loop shows up when the client can't hold a stable connection to OpenAI's backend, which happens during elevated error-rate incidents or when one specific service is degraded while the rest of ChatGPT stays up. Waiting a few minutes and refreshing usually resolves it once the incident clears.
What should I do while ChatGPT is down?
Check status.openai.com for an acknowledged incident, try a different network or device to rule out a local issue, and avoid repeatedly resubmitting the same request. For time-sensitive work, keep a fallback plan rather than waiting indefinitely.
More AI Explainers, Written Clearly
Browse our full library of AI guides, updated as the industry changes.
More from the creator
Explore two other projects from the same creator.