Codex Not Working? Here's Why and How to Fix It (2026 Guide)

Codex not working almost always traces back to one of four causes: you've hit a usage or rate limit, your authentication token expired, the CLI or extension is out of date, or OpenAI is having an outage. This guide is for developers using the Codex CLI, VS Code extension, or ChatGPT-integrated Codex who are seeing stalled responses, auth errors, or "rate limit exceeded" messages.

• Codex usage resets on a rolling 5-hour window for ChatGPT Plus/Pro/Business plans, similar to other LLM coding tools (OpenAI Help Center).
• Most "not working" reports are session/auth issues, not actual outages, per OpenAI's status page history.
• The Codex CLI ships frequent updates, so an outdated binary is a common silent cause of broken behavior (OpenAI Codex docs).

Why is Codex not working?

Codex stops working for one of a handful of reasons: you've exceeded your plan's rate limit, your login session or API key expired, there's a network/proxy block, or OpenAI's backend is degraded. Each has a distinct error signature, which makes it quick to triage once you know what to look for.

If you're on the CLI, run the command with verbose logging (codex --verbose or check the local log file) before assuming it's an outage. Most failures surface a specific HTTP status code (401, 403, 429, or 5xx) that tells you exactly which bucket you're in.

How do you fix Codex when it's not working?

Work through these steps in order, since they're ranked by how often each one is the actual cause:

Symptom	Likely cause	Fix
"Rate limit exceeded" / 429 error	You've hit your 5-hour or weekly usage cap	Wait for the reset window or upgrade plan tier; see our Codex rate limit exceeded guide
401 Unauthorized	Expired login session or revoked API key	Run codex login again or regenerate your API key in the OpenAI dashboard
Hangs with no response	Network/proxy/firewall blocking the API endpoint	Check corporate VPN/proxy settings; test with curl against the OpenAI API directly
Extension doesn't load in IDE	Outdated extension or IDE version mismatch	Update the Codex extension and your IDE to the latest release
5xx server errors	OpenAI-side incident	Check status.openai.com before troubleshooting locally
Context errors / truncated output	Prompt or repo context exceeds the model's context window	Narrow the files/paths you're pointing Codex at, or split the task

Is Codex down right now?

Codex is down only during confirmed OpenAI incidents, which are rare and typically resolved within an hour. Check OpenAI's live status page first: if it shows "operational" for the Codex/API components, the issue is local (auth, network, or rate limit) rather than an outage.

If status.openai.com shows degraded performance, there's nothing to fix on your end beyond waiting; retrying aggressively during an incident can make rate-limit recovery slower once service resumes.

Why does Codex hit rate limits so fast?

Codex rate limits hit quickly because large repo context, multi-file edits, and agentic multi-step tasks consume tokens much faster than a single chat message. A single "fix this bug across the codebase" prompt can burn through a meaningful chunk of a 5-hour window if Codex is reading several files to build context.

OpenAI structures Codex usage around ChatGPT plan tiers (Plus, Pro, Business) rather than pure pay-per-token in most consumer flows, so heavy agentic use on a lower tier exhausts the window faster than typical chat usage (OpenAI Help Center). For a deeper breakdown of limits and reset timing, see our Codex rate limit exceeded article.

How do you check your Codex usage before it breaks?

Check your current Codex usage inside the ChatGPT app under Settings, or via the CLI's session output, which shows remaining quota after each request. Watching usage proactively prevents the "not working" surprise mid-task, since you can see the limit approaching instead of hitting a hard 429 stop.

If you regularly bounce between tools, note that Claude Code has a comparable usage model with its own reset cadence, covered in our Claude Code not working guide, and if you run Claude Code alongside Codex, Usagebar tracks Claude Code's 5-hour and weekly windows directly from your macOS menu bar.

Key takeaways

1. Check the HTTP status code or CLI error message first: 429 means rate limit, 401 means auth, 5xx means OpenAI-side issue.
2. Confirm status.openai.com before assuming it's a local bug.
3. Re-authenticate with codex login if you get 401 errors after a long idle period.
4. Update the Codex CLI/extension regularly; stale binaries cause silent compatibility breaks.
5. Scope your context (fewer files, smaller diffs) to avoid burning rate limit on large agentic tasks.

Sources

• OpenAI Help Center: Using Codex with your ChatGPT plan
• OpenAI Codex developer docs
• OpenAI Status Page