Is Claude down?

status.claude.com is Anthropic’s status page (status.anthropic.com redirects there). It lists claude.ai, the Claude Console, the Claude API, Claude Code, Claude Cowork and Claude for Government as separate components, each marked Operational, Degraded Performance, Partial Outage, Major Outage or Maintenance, with a calendar of past incidents underneath.

If a component is anything but Operational, you have your answer and can stop reading. If everything is green, keep going, because green has a specific meaning: not down. It does not mean not weird.

The right-hand card above is this site’s live reading: today’s 0–100 score, how many people have voted, and how many of those votes landed in the last hour. That last number is the interesting one on a suspected bad day. On a normal afternoon a handful of people wander in; during an incident the rate spikes, the score drops, and the comments fill with the same complaint from different people.

1. Status page, above. Thirty seconds.
2. Try a different product. If the API is fine but claude.ai is not (or the reverse), it is not the model. If Claude Code is failing, try the same prompt in the app.
3. Check your limits. Exhausted usage looks like an outage from inside a session: requests stall, retry or die. Every plan has a meter; look at it.
4. New conversation, same question. A long chat that has stopped responding well is a context problem, not an availability one. In Claude Code, /clear.
5. Then vote. Especially if steps 1–4 came up empty. A score of 12 with “API returning garbage since 14:00 UTC, status page green” is exactly the report the next person needs.

The reason a community signal is worth having next to the official one: both of Anthropic’s published quality incidents happened without a status-page outage. Between August and September 2025, three infrastructure bugs degraded responses for weeks, one of them misrouting up to 16% of Sonnet 4 requests at its peak. Between March and April 2026, three product-layer changes made Claude Code noticeably worse for six weeks while the API stayed healthy. Anthropic documented both, in detail, after the fact.

“Up” and “good” are different measurements. The status page measures the first. The barometer measures the second, badly but continuously, and the history keeps the receipts. For the full record, see Is Claude nerfed?

Slow is not down, and it has boring causes. Under heavy demand, requests queue and sometimes retry; the status page will usually say “Degraded Performance” when that is widespread. At high effort levels the model spends deliberately longer thinking before it answers, which in a terminal can look identical to a hang; in early 2026 Claude Code’s default effort was lowered precisely because long thinking made the UI appear frozen, and then raised again because people preferred the smarter, slower version.

If speed matters more than depth for the task in front of you, lower the effort for that task rather than concluding the service is broken. If it is slow and dumb, that is a vote.