The AI Map Claude Outages
✓ Fact-checked June 24, 2026 Sources: status.anthropic.com · Downdetector · TechRadar · Anthropic Docs

Is Claude Down Right Now? How to Check, What Causes Outages, and What To Do

Whether you are in the middle of an outage right now or planning for reliability, this page tells you exactly how to diagnose the problem, what Claude's error messages actually mean, what caused past outages, and your best options while waiting.

Direct Answer
Go to status.anthropic.com first — it shows real-time status for claude.ai, the API, and Claude for Teams. If status shows operational but you are still getting errors, the problem is likely on your end (VPN, browser cache, extensions). Claude has had 3 notable outages in 2025–2026, including a major June 23, 2026 event with 8,000+ Downdetector reports. Average uptime over 90 days sits at 99.12% — but degraded performance incidents are more frequent than full outages.
Claude “Why Does It...” — Complete Guide Series

The 60-Second Diagnosis — Do This Before Assuming Claude Is Down

Most reported Claude outages are actually local issues — browser cache, VPN interference, or a single session auth problem. Run these steps in order before concluding it is Claude's fault.

1
Check status.anthropic.com
This is Anthropic's official real-time status dashboard. It reports separately on claude.ai (the web app), the Claude API, and Claude for Teams. Understand what each level means: Operational = all systems normal; Degraded Performance = slow or intermittent errors; Partial Outage = some features or regions down; Major Outage = widespread failure. If it shows Operational and you still have issues, move to step 2.
2
Check Downdetector
Go to downdetector.com/status/claude and look at the report graph. A spike in the last 30–60 minutes means other users are experiencing the same thing. Downdetector often picks up incidents before Anthropic updates the status page — during the June 23, 2026 outage, Downdetector hit 8,000+ reports while the status page was still showing “Investigating.”
3
Try incognito / private browsing
Open a new incognito or private window and go to claude.ai. Incognito disables all extensions and uses a fresh cache and session. If Claude works in incognito but not your normal browser, you have either a browser extension conflict (try disabling extensions one by one — ad blockers, VPN extensions, and security extensions are common culprits) or a corrupted browser cache (clear it via Settings › Privacy › Clear browsing data).
4
Try a different browser
If incognito in Chrome fails, try Firefox or Edge. Browser-specific rendering bugs and outdated JavaScript engines can cause Claude’s interface to fail even when Anthropic’s servers are healthy. Claude's web app is a complex React application — some older browser versions or strict security settings cause loading failures that look like outages.
5
Disable your VPN if you are using one
VPN IP addresses get blocked or rate-limited frequently on AI platforms as a bot prevention measure. Your VPN may also be routing you through a region with degraded connectivity to Anthropic's servers. Temporarily disable it and reload claude.ai. If that fixes it, try a different VPN server in your own country, or add claude.ai to your VPN's split-tunnel exclusion list.
6
Try the Claude mobile app
The Claude iOS and Android apps use a different network path than the web app. During partial outages that affect the CDN serving the web app (particularly cloudflare-routed assets), the mobile app often continues working because it communicates directly with the API backend rather than loading through the web CDN. If the mobile app works but the web app doesn’t, this is a web-specific infrastructure issue, not a full Claude outage.
7
Test the API directly (for developers)
If you use the Claude API, run a minimal curl test to isolate whether the problem is the API or just the claude.ai web interface:
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-4-6","max_tokens":10,"messages":[{"role":"user","content":"ping"}]}'

A successful response (HTTP 200) confirms the API is up. An HTTP 529 means the API is overloaded. HTTP 500 or 503 indicates a server-side outage. A connection timeout means network-level routing failure between you and Anthropic.

Quick rule: If 2 of the 3 — status page, Downdetector spike, mobile app failure — all point to Claude, it is a real outage. If only one signals a problem, investigate your local setup before concluding Claude is down.

What the Error Messages Mean — and What to Do

What You SeeWhat It MeansLikely CauseWhat To Do
“Claude is at capacity” or “We’re unable to process your request” Anthropic’s servers are temporarily overloaded — not your personal usage limit Peak-hour traffic surge, viral news coverage, or post-launch traffic spike Wait 5–15 minutes and retry. Try off-peak hours (early morning, late evening US time)
HTTP 529 “Overloaded” (API response) The API endpoint is at capacity and actively rejecting requests Server capacity limit reached at the infrastructure layer Implement exponential backoff with jitter. Start at 1 second, double up to 60 seconds maximum. Do not hammer the endpoint
Connection timeout / page won’t load at all Request never reached Anthropic — or is stuck in routing VPN routing failure, CDN outage, ISP issue, or total infrastructure outage Try step 5 (VPN off) and step 6 (mobile app) from the diagnosis above. Check your internet on another site first
“Something went wrong” (generic error) Catch-all error — could be almost anything Auth session expired, server error, or browser JS failure Reload the page. If that fails, log out and log back in. If still failing, try incognito mode
Infinite loading spinner Claude’s interface loaded but is stuck fetching your session or a response JS/CDN asset failed to load, session auth problem, or streaming response dropped mid-transfer Hard refresh (Ctrl+Shift+R on Windows, Cmd+Shift+R on Mac). If that fails, clear cache and log back in
“Your message was too long” (on a normal-length message) Context window bug — Claude thinks the conversation is longer than it actually is Corrupt conversation state, usually after an interrupted response Start a fresh conversation. If the original conversation was in a Project, it is still saved — just open a new chat and continue from there
“Claude is not available in your region” Your geographic location is restricted from Anthropic’s services Regulatory compliance restriction, not a service outage This is not an outage. Anthropic restricts access in certain countries for legal/regulatory reasons. A VPN may help (check local laws)
HTTP 500 or HTTP 503 Server-side error — Anthropic’s backend is returning an error response Deployment gone wrong, database issue, or full server failure Check status.anthropic.com immediately. This is a real infrastructure problem — nothing you can fix on your end. Wait for Anthropic resolution

Claude’s Outage History — 2025 and 2026

To give you accurate expectations, here is a timeline of notable incidents based on available data from status.anthropic.com, Downdetector, and coverage from outlets including TechRadar.

June 23, 2026 — Major Outage

Most significant incident in 2026: A major outage on June 23, 2026 generated over 8,000 reports on Downdetector within a 24-hour period, making it one of the most widely reported Claude incidents. TechRadar covered the event as it unfolded. claude.ai was unavailable for approximately 40 minutes at peak impact. The Claude API experienced degraded performance — slower responses and elevated error rates — but did not go completely offline. Anthropic issued incident updates via the status page throughout the recovery window. Root cause was not publicly disclosed at the time of writing; the pattern was consistent with a combination of traffic surge and infrastructure failover delay.

March 2026 — API Degradation Incident

An extended degraded performance period in March 2026 affected API response times across several model variants, with Opus-class models most impacted. The incident lasted approximately 2–3 hours at degraded level before recovering. claude.ai continued to function for most users during this period but experienced elevated latency. The March incident coincided with a period of heavy developer adoption following a product announcement, suggesting a traffic-driven overload rather than an infrastructure failure. API users on paid plans received status notifications within 15 minutes of the incident being classified.

December 2025 — Extended Outage

A 2–3 hour outage in December 2025 affected both claude.ai and the API. This was one of the longer single-incident downtimes in Claude’s recent history. Both the consumer product and the API were impacted simultaneously, which is atypical — most incidents affect one more than the other. The timing (US business hours peak) and symmetrical impact suggested an infrastructure-level failure rather than application-layer issue. Anthropic issued a post-incident summary noting infrastructure improvements made as a result.

Recurring Pattern: Elevated Opus 4.8 Error Rate (June 2026)

As of June 24, 2026, the status page shows an elevated error rate specifically on Claude Opus 4.8 that has been active across multiple incidents in June — including a documented service disruption on June 18 (06:55–07:40 UTC) and multiple Opus-specific issues on June 16. This is a model-specific pattern, not a full platform outage. Sonnet 4.6 and Haiku 4.5 have remained more stable during this period.

Uptime in numbers: Anthropic reports approximately 99.12% uptime over the trailing 90-day window. At 99.12% annual uptime, a service experiences roughly 77 hours of downtime per year — about 3.2 days. In practice, this is not 3.2 continuous days; it is spread across dozens of short incidents (30–90 minutes each) plus occasional longer events. Degraded performance incidents — where the service is slow but functional — happen more frequently than full outages and are not always counted in the uptime percentage.

What These Incidents Had in Common

PatternDetail
API vs claude.ai impactIn most incidents, the API degraded but stayed functional while claude.ai went fully down. Invest in API-based access for critical workflows
Peak US hours correlationIncidents disproportionately cluster around 9am–2pm US Eastern — traffic peak drives overload before auto-scaling catches up
Resolution timeMost incidents resolved in 30–90 minutes. Incidents lasting more than 2 hours are exceptions, not the rule
Status page lagAnthropic’s status page typically acknowledges incidents 10–20 minutes after they begin. Downdetector is faster for early detection

Why Claude Goes Down — The 4 Real Causes

1. Traffic Spikes

The most common cause. When Claude is mentioned in viral content, a major tech publication covers a new feature, or Anthropic launches a new model, usage can spike 5–10x within minutes. Auto-scaling on AWS and Google Cloud does not respond instantaneously — there is a lag of 3–10 minutes before new capacity comes online. During that window, servers are overloaded and users see “at capacity” errors.

Typical duration: 15–45 minutes. Status page shows: Degraded Performance or Partial Outage. What you can do: Wait it out. Try again every 10–15 minutes. Early morning retries (before US peak hours) usually work.

2. Infrastructure Failures

Anthropic runs on a combination of AWS and Google Cloud. When an underlying cloud availability zone or specific service component fails (database, load balancer, CDN), the failure can cascade to claude.ai or the API even though the root cause is not Anthropic’s own code. These events are rarer but typically longer-lasting because the fix depends on the cloud provider restoring service.

Typical duration: 30 minutes to 2+ hours depending on the cloud provider’s fix timeline. Status page shows: Partial or Major Outage. What you can do: Check the AWS Service Health Dashboard or Google Cloud Status page to understand scope. Nothing you can do except wait.

3. Deployment Issues

Model version updates, new feature rollouts, and infrastructure changes occasionally introduce unexpected behavior — broken prompts, increased latency, elevated error rates. When Anthropic detects this, they initiate a rollback. During the rollback period (typically 20–60 minutes), service can be intermittent or fully down. Post-incident reviews often show these as the root cause of the more surprising, mid-day outages that do not correlate with obvious traffic spikes.

Typical duration: 20–60 minutes. Status page shows: Investigating, then Identified, then Monitoring. What you can do: The rollback is the fix — follow status page updates.

4. DDoS and Abuse Traffic

Rare but documented across all major AI platforms. A large volume of automated requests from malicious actors can overwhelm rate-limiting infrastructure before blocks are applied. Anthropic does not publicly confirm specific DDoS events for security reasons, but the pattern (sudden onset, uneven geographic impact, rapid recovery after mitigation) is distinct from organic traffic spikes.

Typical duration: 15–45 minutes once mitigation is applied. Status page shows: Usually labeled as “elevated error rate” without specifying cause. What you can do: Nothing on your end. If you are an API user, your legitimate requests will begin succeeding once mitigations are in place.

The Difference Between Claude Down, Degraded, and Partial Outage

Most people searching “is Claude down” are actually experiencing degraded performance, not a full outage. The distinction matters because your response strategy is different.

StateWhat You ExperienceStatus Page LevelAPI StatusBest Action
Total Outage claude.ai cannot be reached at all, or every request returns HTTP 500/503 Major Outage Also down or returning 500s Switch to alternative (ChatGPT, Gemini) and set up an alert for recovery
Degraded Performance Site loads but responses take 30–60s instead of 5–10s; frequent mid-conversation errors; some requests succeed while others fail Degraded Performance Functional but slow; elevated error rate Shorten prompts, avoid heavy tasks, retry failed requests once
Partial Outage Core chat works but specific features fail (Artifacts broken, web search unavailable, file uploads rejected) Partial Outage Usually unaffected Use Claude for text-only tasks until the specific feature recovers
Operational (your issue) Claude seems down but status page shows green; mobile app works fine Operational Operational Run the 7-step diagnostic above — VPN and browser cache are the top suspects
Important nuance: Degraded performance incidents are not counted the same way in uptime calculations. A period where claude.ai is operational but responding in 45 seconds instead of 8 seconds is technically “up” — but it feels broken for interactive use. Check the status page incident history (not just current status) to see if there is a recent degraded performance event that would explain what you are experiencing.

What To Do While Claude Is Down

Option 1 — Try Claude Mobile vs Claude Web

During partial outages, one often works when the other does not. The Claude iOS and Android apps route differently than the web app. If you are stuck on web, open the mobile app. If mobile is also failing, you have a genuine platform outage and need an alternative.

Option 2 — Use the API Directly

The Claude API is architecturally more resilient than claude.ai during partial outages. If you have an Anthropic API key, you can query the API directly even when the web interface is down — as was the case during June 23, 2026. Developers who keep a simple API test script or a fallback integration are able to continue working through most partial outages.

Option 3 — Set Up a Recovery Alert (So You Know the Moment It’s Back)

The most productive response to an outage is to switch tasks, set an alert, and come back when Claude recovers. Options:

Subscribe to Anthropic status emails: Go to status.anthropic.com › click “Subscribe to Updates” › enter your email. You will get email alerts for every incident start and every resolution. This is the most reliable notification method.

Option 4 — Temporary Alternatives During an Outage

AlternativeURLBest For (in Claude’s absence)Gaps vs Claude
ChatGPT (GPT-4o) chat.openai.com Writing, coding, analysis, image generation, browsing Shorter context window on standard tier; different writing style; no equivalent of Claude’s Artifacts for documents
Gemini Advanced gemini.google.com Research with Google search grounding, long document analysis, multimodal tasks Less precise on complex code; different tone; requires Google account
Perplexity perplexity.ai Research, fact-checking, current events — always has live web access Not a general-purpose assistant; limited for creative writing or long-form generation
OpenAI API platform.openai.com Developers: drop-in API fallback using GPT-4o or o3 Different API schema, different model behavior; requires prompt adaptation for best results

Option 5 — For API Developers: Automatic Failover

If you are building a production application on the Claude API, implement automatic failover rather than manual alternatives. A basic pattern:

// Pseudo-code: Claude-first with OpenAI fallback
async function callAI(prompt) {
  try {
    return await claudeAPI(prompt, { timeout: 15000 });
  } catch (err) {
    if (err.status === 529 || err.status === 503 || err.status === 500) {
      console.log('Claude unavailable, failing over to OpenAI...');
      return await openaiAPI(prompt); // fallback
    }
    throw err; // surface other errors (auth, bad request)
  }
}

This pattern ensures your users see zero downtime even during full Claude outages. The key is failing over only on 5xx/529 errors — not on 4xx errors (which indicate a request problem, not a server problem).

Setting Up Alerts — Know Before Your Team Does

1. Anthropic Official Status Page (Email)

Go to status.anthropic.com and click “Subscribe to Updates.” Enter your email. You will receive notifications for incident start, status changes, and resolution. Free. Reliable. This is the first thing every Claude-dependent developer should set up.

2. Downdetector Alert for Claude

Visit downdetector.com/status/claude and create a free alert. Downdetector sends an email when crowd reports spike above a threshold. Because it is crowd-sourced, it sometimes triggers 10–15 minutes before the official status page acknowledges an incident.

3. UptimeRobot or BetterStack (For API Users Who Want Custom Alerts)

Both services offer free tiers with 5-minute check intervals. Set up a monitor pointing to a simple API health check (a low-token request to api.anthropic.com/v1/models using your API key). Configure SMS or Slack notifications. When the API goes down, you are notified within 5 minutes. When it recovers, you are notified again — you can resume your workload without manually checking.

# Monitor endpoint for UptimeRobot / BetterStack
URL: https://api.anthropic.com/v1/models
Method: GET
Header: x-api-key: YOUR_API_KEY
Header: anthropic-version: 2023-06-01
Expected status: 200
Check interval: 5 minutes

4. Status Page RSS Feed

Anthropic’s status page provides an RSS feed at https://status.anthropic.com/history.rss. Add this to any RSS reader (Feedly, Inoreader, or your browser) to get incident updates in your existing news feed workflow without a separate email alert.

Claude API vs claude.ai — Which Is More Reliable?

The short answer: the API is more resilient, and the difference matters most during partial outages.

FactorClaude APIclaude.ai Web App
SLA targetHigher — enterprise contracts include explicit SLA commitmentsLower — consumer product, no SLA guarantee on free or Pro
June 2026 outage behaviorDegraded performance (slow), stayed onlineFully unavailable for ~40 minutes
Infrastructure layerDirect backend connection with retry capabilityWeb CDN + auth layer + React app on top of API — more failure points
Retry capabilityBuilt into your code — exponential backoff means most transient failures resolve automaticallyManual refresh only
Business-critical recommendationUse API with retry logicDo not depend on for production workflows
Enterprise supportDedicated support line, faster incident escalationStandard support queue
For any production application: Always call the Anthropic API directly with retry logic. Never depend on the claude.ai web UI for automated or business-critical workflows. The API’s resilience advantage over the web app is most visible during the type of incident (partial CDN or auth-layer failure) that happens most frequently.

Uptime Comparison — Claude vs ChatGPT vs Gemini vs Perplexity

Platform2025–2026 UptimeNotable IncidentsPaid SLAStatus Page
Claude (claude.ai) ~99.12% June 23 2026 (major, 8K+ reports), Dec 2025 (2–3h), March 2026 (degraded) No published SLA for Pro/consumer status.anthropic.com
Claude API ~99.5%+ Stayed degraded-but-online during June 2026 event; Dec 2025 affected Enterprise: explicit SLA in contract status.anthropic.com
ChatGPT (chat.openai.com) ~99.0% Multiple incidents in 2025; Dec 2024 Christmas outage was widely covered No SLA for Plus/consumer status.openai.com
OpenAI API ~99.4% Multiple 2025 incidents; generally more resilient than ChatGPT UI Enterprise: explicit SLA status.openai.com
Gemini (gemini.google.com) ~99.5% Few major public incidents; benefits from Google’s global infrastructure scale Workspace plans include SLA workspace.google.com/status
Perplexity (perplexity.ai) ~98.8% Periodic slowdowns; smaller infrastructure team means slower incident response No formal SLA for Pro status.perplexity.ai

Uptime figures based on 90-day trailing data from respective status pages as of June 2026. Consumer and API products may have different actual uptime. Uptime percentage alone does not capture incident frequency or severity.

Context on the numbers: The difference between 99.0% and 99.5% uptime sounds small but means 44 hours vs 22 hours of downtime per year. All major AI platforms have meaningful downtime. The real differentiator for production use is not which platform has the highest uptime — it is which one has the best retry semantics and an API-level fallback story.

Frequently Asked Questions

Is Claude down right now? +
Check status.anthropic.com for real-time status. If it shows all systems operational but you cannot access Claude, the issue is likely on your end — try incognito mode, a different browser, or disable your VPN. Claude had a major outage on June 23, 2026 that generated 8,000+ Downdetector reports and affected both claude.ai and the API.
Where can I check if Claude is having an outage? +
Two places: (1) status.anthropic.com — Anthropic's official status page with real-time incident reports for claude.ai, the API, and Claude for Teams; (2) downdetector.com/status/claude — real-time crowd reports from actual users. If Downdetector shows a spike but the status page shows operational, Anthropic is aware and investigating — expect an update within 10–20 minutes.
Why does Claude say "at capacity"? +
The "Claude is at capacity" message means Anthropic's servers are temporarily overloaded — usually during peak US business hours or during a viral traffic spike after a product launch or news coverage. It is not related to your personal usage limit. Wait 5–15 minutes and try again. API users should implement exponential backoff retry logic rather than hammering the endpoint.
How long do Claude outages usually last? +
Most degraded performance incidents last 30–90 minutes. The June 23, 2026 outage — one of the most significant — resulted in claude.ai being unavailable for approximately 40 minutes while the API had degraded but functional performance. Total outages lasting more than 2 hours are rare; Anthropic's 90-day uptime averages around 99.12%, equating to roughly 77 hours of downtime spread across the year.
Why does Claude work on my phone but not my computer? +
This usually indicates a browser-specific or extension-specific problem on your computer rather than a Claude outage. Try your computer in incognito mode — if that works, a browser extension is blocking Claude. Also check if you are on a VPN on your computer but not your phone. Partial outages can affect the web app more than the mobile app due to different infrastructure routing paths.
Is the Claude API more reliable than claude.ai? +
Generally yes, during partial outages. During the June 23, 2026 major incident, the API had degraded performance but did not go fully down — claude.ai was unavailable for approximately 40 minutes. Anthropic maintains higher SLA targets for the API than for the consumer product. Business-critical applications should always use the API with retry logic rather than depending on claude.ai.
What should I use instead of Claude when it's down? +
For most Claude use cases: ChatGPT (chat.openai.com) using GPT-4o covers writing, analysis, and coding. Gemini Advanced (gemini.google.com) is strong for research and Google integration. Perplexity (perplexity.ai) covers research and fact-checking with live web access. None are drop-in replacements for Claude's strengths in long context or nuanced writing, but any will cover a critical deadline in a pinch.
How do I get notified when Claude is back up? +
Three options: (1) Go to status.anthropic.com and click "Subscribe to Updates" for email alerts on any status change; (2) Create a Downdetector alert for "claude" at downdetector.com; (3) API users can set up a free UptimeRobot or BetterStack monitor pointing to api.anthropic.com/v1/models with a 5-minute check interval to get SMS or email notification the moment the API recovers.
🗺

See How Claude Compares to ChatGPT and Gemini

Uptime, pricing, usage limits, and honest verdicts — all verified side by side.

ChatGPT vs Claude — Full Comparison →

More in This Series