Is Grok down?
Open incidents for the xAI API, Grok Build and grok.com from status.x.ai are below. During an outage Grok Build retries on its own, then shows "Service unavailable (503): The service is busy. Wait a minute and send again." or "Server error (500)". Let it retry, try xAI's US endpoint, or switch Grok Build to a custom model from another provider.
Live xAI API and Grok Build status
Checked
No open incidents
- xAI API (api.x.ai)No open incident
- US endpoint (us.api.x.ai)No open incident
- Grok BuildNo open incident
- grok.com (chat app)No open incident
- Last incident: grok.comresolved Oct 7
Source: status.x.ai, xAI's incident feed: it lists incidents, not live uptime, so a row shows any open incident on that service. Read on our server about every 2 minutes.
Outage, or your plan?
| What you see | What it means |
|---|---|
| Service unavailable (503) | "The service is busy. Wait a minute and send again." xAI answered 502 or 503 on every retry. Check the rows above |
| Server error (500) | "Something went wrong on our side. Wait a minute and send again." An internal error at xAI. If it repeats for minutes, expect an incident above |
| Request timed out (504) | "The server took too long to respond. Try again shortly." No answer in time, often during an incident |
| Retrying (attempt 3)... | Shown after the error, as in "Service unavailable (503) | Retrying (attempt 3)...". Grok Build is retrying on its own, up to 15 times by default |
| Connection failed | "Check your network and try again." Grok Build could not reach the endpoint. If the rows above have no incident, check your network, proxy or VPN |
| Rate limited (429) | "You've hit the rate limit for your plan. Try again later." Your plan's or team's limit, not an outage. xAI's own text can replace the first sentence |
| You hit your weekly limit. | HTTP 402: your plan's weekly pool is used up. Buy credits or wait for the reset |
| xAI API: HTTP 429 | "You are sending requests too frequently and have reached the rate limit." Your team's per-model limit, set by spend tier |
| xAI API: HTTP 400 | "Incorrect API key provided." with code invalid-argument: your key. xAI returns 400 here, not 401 |
Grok Build prints the first column, a colon and the quoted text, as in "Rate limited (429): You've hit the rate limit for your plan. Try again later." xAI's error docs list only 4xx codes; during an outage the API answers with 5xx errors or times out.
Five minutes into a Grok outage
- Check the rows above. An incident like the one on September 22, 2026 named the API, grok.com and Grok Build together; others hit only grok.com or Voice.
- Let Grok Build finish its own retries: up to 15, waiting 2 seconds and doubling up to 30. A 429 gets one retry. Resending in a loop adds load.
- If only the global API is failing and you use an
XAI_API_KEY, setGROK_XAI_API_BASE_URL=https://us.api.x.ai/v1. It servesgrok-4.7andgrok-4.6only, at 10% more per token, and the browser login does not use it. - If every Grok model is failing, switch Grok Build to a custom model from another provider with
/model <name>orgrok -m <name>, using the settings below.
Can Grok Build use another provider?
| Route | Settings |
|---|---|
| Another Grok model | Type /model or run grok models. During an xAI outage, other Grok models are usually affected too |
| Standard Compute | In ~/.grok/config.toml add [model.standardcompute] with model = "standardcompute", base_url = "https://api.stdcmpt.com/v1", env_key = "STANDARDCOMPUTE_API_KEY" and api_backend = "chat_completions", then /model standardcompute |
| Any OpenAI-compatible or Anthropic endpoint | The same [model.<id>] table with that provider's base_url, env_key and api_backend (chat_completions, responses or messages) |
| Check what loaded | Run grok inspect, then grok -p "Hello" -m <name> |
Yes: Grok Build accepts custom models, and a model with its own env_key works while you stay signed in to xAI. It does not switch for you when xAI fails; you pick the other model. Its web_search tool runs on a separate model, so set [models] web_search to your custom model or start with --disable-web-search during an outage.
What Standard Compute does when xAI fails
- Every request goes through OpenRouter. Grok 4.3, the one Grok model we serve, has a single provider there: xAI. When xAI's API is down, Grok 4.3 through Standard Compute is down too, with no other host to fail over to.
- Claude, GPT, GLM, DeepSeek and Kimi models on the same key don't touch xAI, and for those a failed provider can fall back to another provider of the same model.
- 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.
- Grok 4.3 has no approved endpoint in Europe or the US, so a key with a Compliance rule does not run it.
The catch
No gateway fixes an outage in Grok Build's own models: your xAI login, its weekly pool and models such as Grok 4.7 run only on xAI. Grok 4.3 through Standard Compute has xAI as its only provider, so an xAI outage stops it too. Switch to another vendor's model until xAI recovers.
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 Grok down or is it just me?
Check the xAI API, US endpoint and Grok Build rows on status.x.ai. If there is no open incident, read the message: "You hit your weekly limit." and a 429 are your plan or team limits, "Incorrect API key provided." is your key, and "Connection failed" with no incident usually means your network, proxy or VPN.
Is grok.com being down the same as the xAI API being down?
Not always. status.x.ai lists grok.com, the API (global and US), Grok Build, the mobile apps and Voice separately. The grok.com outage on October 7, 2026 did not list the API; the September 22 datacenter incident hit the API, grok.com and Grok Build together.
Can Grok Build use another provider when xAI is down?
Yes. Grok Build accepts custom models in ~/.grok/config.toml from any OpenAI-compatible or Anthropic Messages endpoint. It does not fail over by itself: during an outage, switch with /model <name> or grok -m <name>.
What does 'Service unavailable (503): The service is busy' mean in Grok Build?
xAI answered 502 or 503 and Grok Build's retries, up to 15 by default, all failed. Wait a minute and send again, check status.x.ai, or switch to a model from another provider.
Does Standard Compute fail over Grok to another provider?
No. Grok 4.3 has one provider, xAI, so when xAI is down it is down on Standard Compute too. Same-model failover applies to models with more than one provider, such as Claude, GPT, GLM, DeepSeek and Kimi.