The fastest check is Anthropic’s own status page: open status.claude.com and look at the component you actually use — claude.ai for the chat app, Claude API if a tool or integration is failing, Claude Code or Claude Console if those are what broke. If that component shows anything other than Operational, or an incident is listed at the top of the page, the problem is on Anthropic’s side and there is nothing to fix on your device — wait for the incident updates. If every component is Operational and no incident is posted, Claude is almost certainly not down, and the cause is on your side: a usage limit, a capacity message, a stale sign-in or your network. This guide deliberately gives you that checking method rather than a verdict, because Claude’s state changes by the hour and any page that tells you “Claude is down right now” or “Claude is up right now” is describing the moment it was written, not the moment you are reading it.
The 30-second route:
- Open status.claude.com.
- Find your component in the list and read its status word.
- If an incident is open, expand it and read the latest update and its timestamp.
- Only if the page is fully green, run the own-side checks further down this page.
What the official status page shows
status.claude.com is where Anthropic itself posts incidents, updates them while they run, and marks them resolved. It is the only source that can confirm an outage, because it is the only one written by the people fixing it. Everything else — news posts, social media, third-party trackers — is either repeating this page or guessing from user reports.
The page is organised by component, because Claude is not one single service. The components listed include:
- claude.ai — the chat app on the web and the surface the mobile and desktop apps depend on.
- Claude API (api.anthropic.com) — the interface developers and third-party tools call. Your favourite app that “uses Claude” probably talks to this, not to claude.ai.
- Claude Console (platform.claude.com) — the developer console where API keys, usage and billing live.
- Claude Code — Anthropic’s coding tool.
- Claude Cowork — the collaborative agent surface.
This split is the single most useful thing on the page. Components fail independently: the API can be down while the chat app works, and the Console can be degraded while conversations run normally. So “is Claude down?” is really “is the part of Claude I use down?” — and when a third-party app that embeds Claude stops working while claude.ai answers you fine, the API row is the one that explains it. If you are weighing the two big chatbots against each other, our Claude vs ChatGPT comparison covers how the services differ in normal use; status problems are a separate question from which one is better.
Below the components, the page keeps a history: past incidents with their full update timelines, and uptime figures for each component. If the history shows a quiet month while Claude “keeps going down” for you, repeated failures are more likely to be local than a pattern of outages.
How to read the status words and an incident
Each component carries one status at a time. The vocabulary is small, and the differences matter:
| Status | What it means for you |
|---|---|
| Operational | Anthropic reports that component working normally. |
| Degraded Performance | The component is up but impaired — slow responses, intermittent errors, some requests failing. You may get through if you retry; you may not. |
| Partial Outage | Part of that component is unavailable — a feature, a model, a region or a subset of users — while the rest works. |
| Major Outage | The component is broadly down. Retrying is unlikely to help until the incident moves. |
| Under Maintenance | Planned work. Usually announced in advance, and usually brief in effect. |
An open incident has its own timeline, and the labels on that timeline tell you where the fix stands. Investigating means Anthropic has confirmed symptoms and is looking for the cause — this is the earliest stage, and early updates are often vague because the cause is genuinely not yet known. Identified means the cause has been found and a fix is under way. You may also see Monitoring, where a fix has been applied and Anthropic is watching to confirm it held, and finally Resolved, which closes the incident. Each update is timestamped, so read the newest entry, not the headline you saw quoted somewhere an hour ago: an incident that was “investigating” at the top of the hour may be identified or resolved by the time a screenshot of it reaches you.
Two reading habits prevent most confusion. First, check which components the incident names as affected — an incident about a single connector or feature is real, but it does not mean the chat app is down. Second, check the timestamp of the last update: minutes old is a live problem; the same label in a days-old history entry is not.
Downdetector: a useful second signal, not a verdict
If the official page is green but Claude is plainly misbehaving for you, the next stop is Downdetector’s Claude page. Downdetector does not test services itself; it collects reports from users and compares the volume with a baseline. In Downdetector’s own description of its method, it calculates a typical report volume for each service for that time of day from the previous six months, and only flags a possible incident when reports are significantly higher than that baseline; the status is re-evaluated every few minutes.
That method makes Downdetector good at one thing and weak at another. It is good as an early signal: a sudden spike of reports often appears before an official incident is posted. It is weak as proof, for three reasons. Its reports are unverified — a spike can be driven by one region’s internet provider, a popular post telling people to go and report, or confusion with a different service. It cannot tell you which component is failing or why. And a quiet Downdetector page does not clear Anthropic either; a failure affecting only API customers may never produce a consumer report spike. Use the two together: official page green plus a huge, fast-rising spike means “probable real problem, not yet confirmed — wait and recheck”; green plus normal volume means “look at your own side next.”
Not an outage: the look-alikes on your side
Most “Claude is down” moments end here, because several normal behaviours of the service look exactly like downtime from the user’s chair.
Usage limits. Paid Claude plans run on usage limits — a session window and a weekly allowance — and when one is spent, Claude stops answering and shows a limit message naming the reset time. Nothing is broken, the status page stays green, and no amount of reloading changes it: access returns at the stated reset, or sooner if your plan offers usage credits or an upgrade. This is an account state, not an outage. The limits are part of how the service manages demand — the same demand-management logic behind what a large language model is being expensive to run at scale.
Capacity messages. At peak demand you may see a message along the lines of Claude being unable to respond due to unexpected capacity constraints, asking you to try again soon. Anthropic treats this as load management, not an outage: it is brief, it affects responses rather than the whole service, and it does not appear as an incident on the status page. The correct response is to wait a few minutes and retry, not to reinstall anything.
Sign-in and session faults. A login loop, a “there was an error logging you in” message, or being bounced back to the sign-in page while the status page is green points at your browser’s stored state. Anthropic’s own guidance for sign-in trouble after a confirmation email is to clear the browser’s cache and cookies and try again. The fuller ladder: try a private or incognito window (which rules out extensions and stored cookies in one step), turn off a VPN if you use one, try the mobile app or a different browser, and sign out and back in. If Claude works in the app but not your browser, the service was never down — your browser session was.
App versus web. The mobile and desktop apps and claude.ai in a browser are different doors into the same service, and they do fail separately at the level of your device: an outdated app version, a stalled app cache or a phone network problem can break one door while the other works. Testing the second door is the quickest way to separate “Claude is down” from “this app on this phone is stuck.”
API error codes (for developers). Two API errors look like downtime and usually are not an outage verdict either. A 429 (rate_limit_error) means your own organisation went over its rate limits — the fix is to wait the retry-after interval, not to report an outage. A 529 (overloaded_error) means the API is temporarily at capacity across users; back off and retry, and check the Claude API component on the status page to see whether a confirmed incident is also open.
Your network and region. A workplace or school network, a VPN exit in another country, or a filtered connection can block or degrade the path to Claude while the service itself is healthy. Switching to a phone hotspot for one test answers this definitively: if Claude loads on mobile data, the service and your account are fine, and the fault is in the network you started on.
What to do while a real outage runs
When the status page does confirm an incident on your component, the productive moves are few, and they are mostly about not making things worse:
- Stop troubleshooting your own setup. Do not clear caches, change passwords or reinstall apps in response to a confirmed provider outage; you will only create second problems to untangle afterwards, and none of it shortens the incident.
- Watch the incident, not social media. The incident timeline on status.claude.com is where the resolution will be announced first and most accurately. Reload it periodically rather than following rumour threads.
- Protect your work. If you were mid-conversation, your chat history is stored on Anthropic’s side and normally reappears when service returns; a failed generation does not delete a conversation. Long, unsaved drafts typed into the message box are the vulnerable thing — if you are writing something long during instability, keep a copy in a notes app.
- Use the time or switch tools. For urgent work, another assistant can cover the gap — the trade-offs are covered in Claude vs ChatGPT — but expect your Claude-specific history and memory not to carry over.
- Recheck before resuming heavy use. Wait for the incident to reach Resolved, not just for one message to succeed; degraded services often flap — working for a minute, failing again — before they stabilise.
If the status page insists everything is operational while Claude fails for you across devices and networks for an extended period, that is the point to contact Anthropic rather than keep guessing: help and support routes start from Anthropic’s support site, and reporting the exact error text, the component, and the times you tried is what makes a report actionable.
Frequently asked questions
Where is Claude’s official status page? status.claude.com. It lists each Claude component separately — claude.ai, the API, the Console, Claude Code and Cowork — with current statuses, any open incident with a running timeline of updates, and past incidents with uptime history.
Claude is not working for me, but the status page says Operational. Why? Because most failures are local or account-level: a spent usage limit, a brief capacity limit at peak demand, a corrupted sign-in session, an outdated app, or a network or VPN problem. Work through the own-side checks above — private window, second device or app, different network — before concluding anything about the service.
What is the difference between Degraded Performance and Partial Outage? Degraded Performance means the component works but impaired — slower, with some requests failing. Partial Outage means a slice of it — a feature, model, region or group of users — is unavailable while the rest runs. In practice both mean “retry may work, and the incident timeline matters more than the label.”
Why do different status-checker sites disagree about whether Claude is down? Because almost all of them are mirrors, not witnesses. One tracker re-polls Anthropic’s status feed every minute or so; another runs its own reachability pings, which some Claude surfaces refuse; a third counts user reports. On the same afternoon they can honestly show different verdicts — during the Microsoft 365 connector incident on 12 November 2026, trackers simultaneously showed a partial outage and “No, it’s up,” because the incident hit one connector while the chat app and the API kept working. When verdicts conflict, the component list on the official page is the tiebreaker.
Does a capacity message mean Claude is down? No. Capacity messages are Anthropic shedding load at peak times. They are temporary, they do not appear as incidents on the status page, and the fix is to wait a few minutes and try again.
How long do Claude outages usually last? There is no reliable general answer, and this article will not invent one. The honest source is the incident itself: read its timeline on the status page, where the progression from Investigating to Identified to Resolved tells you more than any average quoted elsewhere.
This article is a checking guide, checked in November 2026 against Anthropic’s official status page and support site linked above. It intentionally makes no claim about whether Claude is up or down on the day you read it — always take the current answer from status.claude.com itself.