Is Codex (OpenAI) down right now?
Live status for chatgpt.com isn't set up in this environment yet - check any URL manually instead.
Checking chatgpt.com right now
A fresh check, started when you opened this page - each probe runs its own independent DNS, TCP, TLS and HTTP request. Anonymous runs sample the network; a signed-in account checks from more locations.
About Codex (OpenAI) outages
Written by the HostTracker monitoring team · updated from live probe data.
Codex is OpenAI's coding agent, and it doesn't run as a separate standalone service: the version built into ChatGPT lives inside chatgpt.com, and the Codex CLI that runs from a terminal authenticates against and sends its requests through the OpenAI API. That means Codex doesn't fail on its own infrastructure so much as it inherits problems from whichever surface it's riding on at the time - a ChatGPT-wide issue affects the in-app agent, while an API-side issue affects the CLI. If a check here or on /en/is-chatgpt-down or /en/is-openai-down shows a problem, that's very likely Codex's problem too, since it has no independent backend of its own to fall back on.
Common causes of Codex (OpenAI) downtime
Task queue backlogs, where a coding task is accepted but sits waiting far longer than usual before Codex actually starts working on it
Sandbox or container start failures, where the isolated environment Codex runs a task in fails to provision, blocking the task before any code even executes
GitHub connection errors that break Codex's ability to read a repository, open a pull request, or push changes, independent of whether Codex itself is reachable
Model capacity constraints during high-demand periods, which slow or queue task completion the same way they would for any other product built on the same underlying models
CLI authentication failures, where a stale or expired login blocks the terminal client from reaching the API even though the API itself is healthy
Not sure whether it's Codex (OpenAI) or your own connection? Read the full breakdown in our website down checker guide for a general walkthrough of diagnosing any outage, from DNS failures to expired certificates.
Other services we track
HostTracker has monitored websites and online services since 2004 and is trusted by 500,000+ sites today, covering 14 check types from 300+ global locations. Want the same multi-location confirmation and instant alerts for your own site or any service you depend on? Getting started costs nothing - start a free monitor or compare plans and pricing.
Questions people ask during an outage
The live probe run on this page tests chatgpt.com from a broad sample of HostTracker's 300+-location monitoring network - many independent networks at the same moment, rather than just your own connection. If every location fails to reach it, the outage is real and affects everyone; if only a few locations fail while most load normally, the problem is local to those regions or networks rather than Codex (OpenAI) being fully down. A single refresh from your own browser can never make that distinction, because it only ever tests from one vantage point on one network.
Yes. HostTracker lets you add any URL - including third-party services you depend on, not just your own website - as a monitored target, and notifies you automatically the moment it becomes unreachable. Before sending an alert, HostTracker re-verifies a failure from multiple locations in its 300+-location network, so you're notified about real, confirmed outages rather than a single location's momentary blip. The free plan checks two monitors every 30 minutes indefinitely at no cost; paid plans check as often as once a minute, with alerts available through SMS, email, voice call, and several messenger apps.
Run the live check at the top of this page - it tests chatgpt.com from HostTracker's 300+ monitoring locations at once and tells you within seconds whether Codex (OpenAI) is reachable for everyone or only failing for you.
Run the live check at the top of this page - it tests chatgpt.com from HostTracker's 300+ monitoring locations at once and tells you within seconds whether Codex (OpenAI) is reachable for everyone or only failing for you.
It can be, though a queue backlog is a narrower symptom than a full outage. Codex tasks go through a processing queue before they actually start running, and a backlog there delays a task's start without necessarily meaning the underlying service is unreachable. If tasks eventually start and complete, just slowly, that points at a temporary capacity crunch; if they never start at all, checking chatgpt.com and the OpenAI API's status separately narrows down which layer is actually failing.
Usually not. A CLI authentication failure most often comes from a stale or expired login rather than the OpenAI API being unreachable, since the CLI needs a valid session to send requests through the API in the first place. Signing out and back in through the CLI resolves most of these. If sign-in fails for many people at once and the API check on /en/is-openai-down also shows a problem, that points at a genuine authentication-service incident instead.
That's typically a GitHub connection problem rather than a Codex or OpenAI issue. Codex needs to reach GitHub separately from generating the code itself, so a connection error, a permissions problem, or an incident on GitHub's side can block the final step - opening a PR, pushing a branch - after the actual coding work has already completed successfully. Checking GitHub's own reachability alongside this page is worth doing when the failure looks specific to the repository step.
Monitor your website from 300+ locations
Start a free trial and get instant alerts when your site goes down - verified from multiple locations before you're notified.