Is Claude AI down?
Check status.claude.com — Anthropic's official status page, updated in real time. It tracks six components separately: claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, and Claude for Government. Over the trailing 90 days as of 2026-07, claude.ai posted 99.47% uptime and the API 99.58%.
Why — the first-principles explanation
The single most useful thing to know: Claude is not one system, so "is Claude down" is usually the wrong question. Anthropic's status page breaks the service into six independently monitored components — claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, and Claude for Government. They fail independently. A real July 14, 2026 incident degraded document creation in claude.ai, Cowork Remote, and Claude Code while other services stayed up. So "Claude is down" from a stranger on social media tells you almost nothing about your specific problem.
There's a second layer most outage trackers miss. Anthropic reports status per model too — incidents have been logged as elevated errors on specific models (Opus, Sonnet, Fable, Mythos, Haiku) rather than on the platform as a whole. That is a direct consequence of how these systems are built: each model is served by its own fleet of accelerators. Fable 5 capacity can be saturated or pulled while Haiku 4.5 hums along untouched. Switching models is a genuine workaround that has nothing to do with your internet connection.
Why do outages cluster at all? Because inference capacity is hard-limited by physical hardware. A web app can absorb a traffic spike with cheap servers. An AI lab cannot conjure more accelerators on a Tuesday afternoon. When demand exceeds the fleet, you get exactly what the status page shows: elevated errors and degraded performance rather than a clean on/off. This is also why Anthropic sells "priority access during high traffic" as a paid Max feature — congestion is the product constraint, not an accident.
One more distinction worth holding: degraded is not down. Most incidents in the record resolve as slowness or elevated error rates, not total blackouts. If you get a response and it's slow, the system is working as designed under load.
An example that makes it click
Think of a hospital, not a light switch. When someone says "is the hospital down?", the answer is almost never yes or no — the ER might be swamped while radiology is empty and the pharmacy is fine.
Claude works the same way. On July 14, 2026, document creation broke while ordinary chat kept running. That's the radiology machine being out, not the hospital closing. And the fix is often the same as at a hospital: if the specialist is backed up, see a different one. Switch from Fable 5 to Sonnet 5 and you're frequently back to work in ten seconds.
How to do it
- Open status.claude.com — this is Anthropic's official real-time status page and the only authoritative source.
- Find your specific component: claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, or Claude for Government. They fail independently.
- Check whether the incident names a specific model. If Fable 5 or Opus is degraded, switch to Sonnet 5 or Haiku 4.5 and try again.
- Subscribe to updates on the status page so you get notified when the incident resolves instead of refreshing.
- If the status page shows all systems operational, the problem is likely on your end — hard-refresh, clear cookies, disable extensions, or try a different browser or network.
- Still broken with a green status page? Contact support at support.claude.com.
Key facts
- Anthropic's official status page is status.claude.com and monitors six components: claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, and Claude for Government.
- Trailing 90-day uptime as of 2026-07: claude.ai 99.47%, Claude Console 99.74%, Claude API 99.58%, Claude for Government 99.93%.
- 99.47% uptime equals roughly 3.9 hours of downtime per month — outages are normal and expected, not rare.
- A partial outage on July 14, 2026 affected document creation in claude.ai, Cowork Remote, and Claude Code while other components remained available.
- Incidents are frequently logged against individual models (Opus, Sonnet, Fable, Mythos, Haiku) rather than the whole platform.
- Anthropic sells 'priority access during high traffic' as a Claude Max feature, confirming congestion is a known capacity constraint.
Anthropic's assistant, known for long-document reasoning and careful, safe answers.
▶ The 60-second explainer (script)
Is Claude down? Go to status dot claude dot com. That's Anthropic's official status page, updated live. But here's the thing most people miss. Claude isn't one system. That status page tracks six separate components: claude dot ai, the Console, the API, Claude Code, Cowork, and Claude for Government. They break independently. On July fourteenth, 2026, document creation went down while regular chat kept working fine. So asking is Claude down is like asking is the hospital down. Wrong question. Ask which part. And there's a second layer: Anthropic logs outages per model. Fable 5 can be throwing errors while Haiku 4.5 is perfectly healthy. Switching models is a real fix, not a superstition. Now the numbers. Over the last ninety days, claude dot ai ran at ninety-nine point four seven percent uptime. Sounds great. It's about four hours of downtime a month. Outages aren't a scandal, they're a Tuesday. If the status page is all green and you're still stuck, it's you. Hard refresh, kill your extensions, try another network. Then contact support.
What authoritative sources say
People also ask
Where is the official Claude status page?
status.claude.com. It is run by Anthropic, updates in real time, and lets you subscribe to incident notifications.
Claude is slow but not dead. Is that an outage?
Usually it's logged as 'degraded performance' rather than an outage. Most incidents in Anthropic's record are elevated errors or slowness under load, not total blackouts.
Why does one model work when another fails?
Each model runs on its own serving capacity. Anthropic logs incidents against individual models, so Fable 5 can be degraded while Sonnet 5 or Haiku 4.5 works normally.
The status page says everything is fine but Claude won't load for me.
That points to a local issue. Hard-refresh the page, clear cookies, disable browser extensions, or try a different browser or network before contacting support.
How often does Claude actually go down?
claude.ai ran at 99.47% uptime over the trailing 90 days as of 2026-07 — roughly four hours of unavailability per month.
The same question, asked other ways
- Is Claude AI down right now?720/mo