Grok and xAI outages: what the status pages actually tell you
What status.x.ai reports, what Grok's own FAQ leaves out, and how to tell an xAI outage from a limit on your own account.
Marta's evening comes apart at about nine, halfway through a long back and forth about a rental contract she does not fully understand. The answer never arrives, sending the message again produces the same nothing, so she does what everybody does and opens another tab to ask whether Grok is down.
A minute later she has xAI's status page and Grok's help pages open, and the honest summary of both is that neither one is about her. Here is what those surfaces settle and what they leave alone, checked on 2 October 2026.
What status.x.ai actually reports
The browser tab says SpaceXAI Status while the page itself is headed Service Status, a jolt if you arrived by typing xai status, though the components underneath carry the familiar names. status.grok.com serves the same page, so the two addresses people trade in forum threads are one surface.
At the top it reported no incidents declared, then one sentence:
We are not actively mitigating any known incidents at this time.
That is narrower than it looks. It describes what xAI has declared, and declaring an incident is a deliberate act that follows the first failures rather than arriving with them. Marta's message had failed twice already.
Below sits a block the page calls live service data, described there as monitoring data that indicates service health even when no incident has been declared, and it is the more useful half. It grids endpoints by source region across eu-west-1, us-central-1, us-east-1 and us-west-2, each pairing carrying a reading for inference and one for non-inference, with a line saying how recently they refreshed. We are not reprinting the percentages, which are a live measurement.
Then the services list, fourteen entries when we looked, each marked available and each with a page of its own: the iOS, Android and web apps, Grok Build, the Office and Workspace plugins, single sign-on, four API endpoints, the API console, the docs, the xAI website and Grok in X. Those component pages carry more than the overview. Grok Build's page reported the service fully operational, and showed a thirty day uptime reading, an issue count sorted into the three classes the site calls outage, disruption and info, and a list of past issues with a duration on each. Incidents also publish to an RSS feed at status.x.ai/feed.xml, each tagged with its component in square brackets, which is why outside trackers, ours included, show those bracketed names.
A section headed Models sits beneath the services list, and the text-only rendering fetched on 2 October 2026 returned nothing under it, so we will not say what it lists. What the page carries nowhere is you: no field for your account, your plan or your remaining usage.
MultiChats puts the leading AI models in one app. Try it free
The API and the app are two separate questions
Two of the searches that land here want different pages. A search for grok api status is about api.x.ai, the endpoint other companies' software calls. A search for grok server status almost always means the consumer app: grok.com in a browser, the phone apps, or Grok inside X. The page keeps them apart with separate components, and a reader who has not noticed the split can read a healthy API entry as reassurance about an app that is failing, or the reverse.
The split has a consequence in our own app, and it cuts against us. When you pick Grok in MultiChats you are on the API side of it: we call xAI's API rather than grok.com. If the consumer components are healthy and a regional endpoint is struggling, Grok can fail for you here while it works in a browser for the person next to you.
Grok Build has a component of its own, which answers the searches asking whether Grok Build is down. xAI's developer documentation describes it as agentic coding on the API and publishes no availability history of its own, so of those two developer surfaces only the status page answers an availability question.
Grok in X is the awkward one. It is listed as an xAI component, yet xAI's consumer FAQ says of it: "While xAI provides the Grok in X service on X.com and X apps, it does not have operational oversight of X's service." Two organisations, one chat box, and the FAQ sends you to X's own Help Center for it.
Grok's own help pages stop short of outages
The consumer FAQ at grok.com/faq, dated 27 August 2026 on the page itself, covers the website and the phone apps: billing, usage and limits, accounts and login, image and video generation, products and models, files and data, teams, and how to reach a human. What it does not contain is a single entry about outages.
It is precise about the account side instead, the half Marta needed. Paid usage runs as one weekly pool shared across products, shown in Settings then Usage as a percentage with a reset date. When the pool is spent, the FAQ says paid features pause until the weekly limit resets, though it calls that no hard stop, and free tier access to Chat and Voice continues on its own schedule.
Nobody joins the two halves up for you, and a chat that will not answer looks the same whether a region is degraded or your weekly pool ran dry twenty minutes ago.
How the same failure reads inside MultiChats
When a model fails in MultiChats the raw upstream error never reaches you: it is classified on the server and replaced with one short sentence about the model you chose, since upstream errors arrive carrying host names and dashboard links with no business on your screen.
The sentences differ by class, deliberately. A provider that has gone unavailable, or a credential problem at our end, produces "Grok 4.3 isn't available right now. Please try a different model." A provider that throttled us, or one that never produced a first token, produces "Grok 4.3 isn't responding right now. Please try again in a moment." That second wording replaced one that called the model busy, because busy reads as our servers straining when the straining is somewhere else.
The failures that are the provider's to fix also add a Check status action alongside that sentence, which opens our status page in a new tab. Your own limit never does, and it never appears as a model error either. It arrives as a banner above the composer that counts down the messages you have left, then reads "You have no messages left." on a free account, which gets 50 messages every 24 hours. For a subscriber whose allowance is spent it opens "You've used all your messages", then says when they reset or what the tier above would add. Both Grok entries are paid, so that is the banner a Grok user meets; the phone apps head both versions "No messages remaining".
So a sentence naming the model is about the provider, and a banner counting messages is about you. Our own page adds what xAI's cannot, which is what our requests did. It lists a model only when requests we sent in the last thirty minutes failed, which can happen while its provider reports everything as normal.
What to do while Grok is unavailable
Switch model inside the same thread. That is the answer to most of these evenings, and what it preserves matters: the conversation stays where it is, history intact, and the next answer comes from a different model. Branching opens a new thread from the point you chose, and there is no side-by-side view in MultiChats, so switching is what keeps you in the conversation you were having.
The two xAI entries are Grok 4.3 and Grok 4.3 Reasoning. They share a provider, so they tend to have their bad hours together, and the useful choices in that hour are the other 58 models, from the other 17 providers. Pick on what the task needs, and if the thread has grown long, pick something with room for it.
One thing worth knowing as you move between models. We do not train on your conversations. What the provider behind a model does with them varies, and the models where we know of a difference carry a notice on their info card in the picker. Both Grok entries carry one, reading in full: "xAI may use your conversations to train Grok models." Most entries carry no notice, so an empty card says nothing either way.
What a status page can settle, and what it cannot
Marta's evening ended in the dull way these do. Nothing had been declared, the regional readings looked ordinary, her account was in good standing, and the model still would not answer. She switched models in the same thread, got through the clause she was stuck on, and found Grok answering again the next time she opened it.
The judgement to take from it is that a status page answers a question about the provider's fleet, and is good at that, while the question you have is why this reply has not come. Yours needs the other half: your plan and your allowance first, then the app, then the surface you are using it on. That order saves the twenty minutes you would spend refreshing a page that has already told you what it knows. Our Grok status page reads xAI's feed and puts the current state and recent incidents on one screen, which is the whole job a status page can honestly do. You will still need to check your account and app.