Is Codex down?
The Codex rows from status.openai.com are below. During an outage Codex shows "Reconnecting... 1/5", then "stream disconnected before completion", "We’re currently experiencing high demand" or "Selected model is at capacity". Let it retry, switch model with /model, or run Codex CLI on another provider.
Live Codex status
Checked
All systems operational
- CLIOperational
- VS Code extensionOperational
- Codex WebOperational
- Codex APIOperational
- ResponsesOperational
- Last incident: Degraded performance in Codex and ChatGPT Workresolved Oct 14
Source: status.openai.com, Codex rows and Codex incidents only. Read on our server about every 2 minutes.
Outage, or your plan?
| What Codex shows | What it means |
|---|---|
Reconnecting... 1/5 | The response stream dropped and Codex is retrying on its own, five times by default |
| "stream disconnected before completion" | The stream broke before the answer finished. If the Codex rows above are green and it keeps happening, check your network, proxy or VPN |
| "We’re currently experiencing high demand" | "…which may cause temporary errors." Codex's wording for an HTTP 500 from OpenAI. Retry, and watch for an incident above |
| "Selected model is at capacity" | "Please try a different model." HTTP 503 server_is_overloaded: that model is full. Type /model and pick another one |
| "exceeded retry limit" | Codex gave up after its retries. The last status after it says why |
| "You’ve hit your usage limit" | Your ChatGPT plan's window, not an outage. See the Codex usage limit fix |
| "Quota exceeded" | "Check your plan and billing details." Billing on your OpenAI API key |
| "Your access token could not be refreshed" | Your login expired. Log out and sign in again |
Five minutes into a Codex outage
- Check the CLI, VS Code extension, Codex Web and Codex API rows above. An incident that names Codex is listed with them.
- Let Codex finish its own retries. Restarting it in a loop adds load and loses the turn.
- If one model is at capacity, type
/modeland pick another one, or start withcodex --model <id>. - If OpenAI is failing for every model, run Codex CLI on a provider that serves other vendors' models, with the settings below.
Run Codex CLI on another provider
| Route | Settings |
|---|---|
| Another OpenAI model | Type /model, or set model = "…" in ~/.codex/config.toml |
| Standard Compute, one command | Run standardcompute launch codex and approve the browser connection. Your normal Codex settings are kept |
| Standard Compute, by hand | In ~/.codex/config.toml add [model_providers.standardcompute] with name = "Standard Compute", base_url = "https://api.stdcmpt.com/v1", env_key = "STANDARD_COMPUTE_API_KEY" and wire_api = "responses", then set model_provider = "standardcompute" |
| A non-OpenAI model | On that provider, set the model to another vendor's ID, such as anthropic/claude-sonnet-5.5 or z-ai/glm-5.3 |
A custom provider must serve the OpenAI Responses API: Codex no longer accepts wire_api = "chat". Codex Web tasks and the ChatGPT login stay on OpenAI.
What Standard Compute does when OpenAI fails
- Every request goes through OpenRouter. GPT requests go to OpenAI first; if OpenAI fails, OpenRouter can send the request to another provider of the same GPT model, such as Azure.
- Each failed call is retried twice on a 429, a 5xx or a network error, after 0.5 and then 1 second, or after the provider's Retry-After up to 8 seconds. If every attempt fails, your agent gets HTTP 503 with "The model is temporarily unavailable. Please retry shortly."
- A specific model ID always stays that model: failover changes the provider, never the model.
- Codex on a Claude, GLM or DeepSeek model through the same key does not touch OpenAI at all.
The catch
No gateway fixes an outage in Codex itself: the ChatGPT login, your plan's usage and Codex Web tasks run only on OpenAI. And when a GPT model fails at OpenAI and Azure together, every provider of that model is affected, Standard Compute included; switch to another vendor's model or wait.
Where Standard Compute fits
Standard Compute is built to make coding agents cost less: its router sends each step to the cheapest model that can do it, on providers we vet for price, uptime and speed, and a failed provider falls back automatically to another provider of the same model. An opt-in Compliance rule keeps a key, a team or the whole company on endpoints in Europe, the United States or both, and we don't train on your prompts or store them. One key at https://api.stdcmpt.com/v1 serves Claude, GPT, Gemini, DeepSeek, GLM and Kimi models, from $19 a month (USD).
FAQ
Is Codex down or is it just me?
Check the Codex rows on status.openai.com: CLI, VS Code extension, Codex Web and Codex API. If they are green, read the message: "You’ve hit your usage limit" is your plan, a token-refresh error means sign in again, and repeated "stream disconnected before completion" on green rows usually means your network or proxy.
What does 'Reconnecting... 1/5' mean in Codex?
Codex lost the response stream and is retrying on its own, up to five times by default (stream_max_retries in config.toml). If the last attempt fails, the error after it says why.
Is ChatGPT being down the same as Codex being down?
Not always. Codex has its own rows on status.openai.com, and an incident can hit ChatGPT without Codex or the other way round. Codex with a ChatGPT login still needs that account to work.
Can I keep using Codex CLI while OpenAI is down?
Yes, through a custom provider that serves the Responses API and a model from another vendor. On Standard Compute, run standardcompute launch codex and use a model such as anthropic/claude-sonnet-5.5. Codex Web tasks still need OpenAI.