Is Claude Down Right Now? 3 Ways to Check in 30 Seconds, and 5 Things to Do While You Wait
Table of Contents
Short answer: open status.claude.com. It lists claude.ai, Claude API, Claude Code and Console as separate components; whichever is red is the one that is broken. If everything is green and you still cannot use it, nine times out of ten it is not an outage: you have hit your usage limit, or the model you are on is temporarily overloaded and switching models gets you moving again.
This article is short on purpose, because you are probably stuck right now. Check first, then decide whether to wait or route around.
First, tell the three "looks like an outage" cases apart
| What you see | What it is | What to do |
|---|---|---|
| Spinner forever, API returns 5xx, status page shows red | A real outage | See "5 things to do" below |
| A message about usage limit, quota, reset time | Your own usage cap | Wait for the reset, or read Claude off-peak: double your usage |
Claude Code says 529, overloaded, Overloaded |
One model is overloaded right now | /model to switch; it usually clears in minutes |
Usage limits get confused with outages the most. The tell: an outage hits everyone and shows on the status page; a usage limit hits only you and the status page stays green.
3 ways to check in 30 seconds
1. The status page: status.claude.com (the old status.anthropic.com redirects there). The front page lists six components: claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, Claude for Government. Look at the one you use.
2. One command (in a terminal, handy for Claude Code users):
curl -s https://status.claude.com/api/v2/status.json | grep -o '"description":"[^"]*"'
All Systems Operational means the platform is fine. Per component:
curl -s https://status.claude.com/api/v2/components.json
3. See whether others are down too: search X for "Claude down" and look at the last ten minutes. The status page usually lags users by a few minutes; this fills the gap.
Real numbers: how often did it go down in the last 30 days
On 23 August 2026 we pulled the incident history from the status page's public API for 24 July to 23 August:
- 30 incidents: 17 minor, 9 major, 2 critical, plus 2 no-impact maintenance notices
- Median 63 minutes from creation to resolution; a quarter resolved within 34 minutes, three quarters within 93
- The longest was about 7 hours
So "Claude is down again" is not your imagination; it averages one incident a day. It is also not the end of the world; most are over within the hour. That frequency drives the advice below: do not switch tools every time, have a routine you follow.
5 things to do right now
1. Check which component is down. claude.ai being down does not mean the API is down; Claude Code being down does not mean the web app is. The status page separates them, and switching entry points often gets you going.
2. If one model is overloaded, switch models. In Claude Code type /model and move from Opus to Sonnet or the other way. Many incident entries read "degraded performance for model X", not everything at once.
3. Use a different infrastructure. Claude is also served on AWS Bedrock and Google Vertex AI, which are separate data centers. Claude Code switches with an environment variable (CLAUDE_CODE_USE_BEDROCK=1 or CLAUDE_CODE_USE_VERTEX=1). You need the accounts and keys set up beforehand; applying during an outage is too late. If Claude is your production line, it is worth having.
4. Carry your project context to another tool for a while. If your CLAUDE.md is well written, hand it to Gemini CLI or Codex as the system prompt and keep doing the non-urgent work. Not a long-term replacement, just a tourniquet.
5. Commit, write down the next step, go do something else. Median 63 minutes is exactly long enough for email, a meeting, or a walk. Before you leave, put "where I am and what is next" in a NOTES.md so you do not have to reconstruct it when you come back.
If you run automation on Claude, this is different
Everything above is for a human at the keyboard. If Claude runs your schedules, agents, or customer replies, one incident a day is a daily risk you cannot rely on a human to notice. We once ran this setup 60 days without an interruption (an early-June milestone; we have hit new kinds of silent failure since and added more monitoring layers). The core design: every call has a timeout, transient errors retry with backoff, permanent errors page a human immediately, and monitoring distinguishes "the service is alive" from "the service is actually doing work". Details in Silent failure detection for AI automation, and the Windows-specific toolkit in Three rough edges of running Claude Code + Telegram on Windows.
Incident statistics come from the public status.claude.com API (/api/v2/incidents.json), computed on 23 August 2026 for the window 24 July to 23 August. The numbers move; pull them yourself for the latest.
FAQ
Is Claude down right now? How do I check?
Open status.claude.com. It lists claude.ai, Claude API, Claude Code and other components with their own status. Or run curl -s https://status.claude.com/api/v2/status.json and read status.description. If everything is green and you still cannot use it, you have most likely hit your usage limit or a single model is overloaded; that is not a platform outage.
How long do Claude outages usually last?
Across the 30 incidents on the status page between 24 July and 23 August 2026 (including 2 no-impact maintenance notices), the median time from creation to resolution was 63 minutes; a quarter resolved within 34 minutes and the longest took about 7 hours. Most of the time, doing something else for half an hour is the right move.
Claude Code shows 529 or overloaded. Is that an outage?
Not exactly. 529 means that particular model is overloaded right now, usually for a few minutes, and often only one model is affected. Switch with /model and you can normally keep going. If it persists for more than ten or fifteen minutes, check the status page.