Don't stop here
Hand-picked guides our readers explore right after this one.
Expert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read the guideStunning image generation with Midjourney prompt mastery
Read the guideAI prompts for code generation, debugging, architecture, DevOps, and career growth
Read the guideWhen Claude stops working, I start by identifying which surface failed: claude.ai, the mobile app, Claude Code, or an application using Anthropic's API. Then I check status.anthropic.com before changing anything. A 529 means Anthropic is overloaded, a 500 is a server-side API error, a 504 is a gateway timeout, and a 400 means the request itself is invalid. A green status page points me toward a local session, browser, network, billing, permission, usage, or request-size problem. This guide uses that distinction to help you choose the right fix instead of repeatedly refreshing and hoping.
Claude down and need to keep building?
If you were using Claude to build a website or app, you don't have to wait out the outage. Lovable is a separate AI app builder that turns a prompt into a working, deployed full-stack app — so you can keep shipping. It's free to start.
Affiliate link, we may earn a commission at no extra cost to you.
claude.ai shows a blank page or infinite spinner on load
Messages send but no response appears, cursor just blinks
'Error sending message' or 'Something went wrong. Please try again.'
'Overloaded' / 'Claude is experiencing high demand right now' (API error 529)
API error 500 ('Internal server error') from the API or an app built on Claude
Claude Code shows 'API Error: 400' or 'API Error: 529' in the terminal
Claude Code feels slow or gives worse answers — degraded performance today
'This model isn't available right now. You can switch to another model to continue using Claude.'
'Claude couldn't finish this response. Try again in a moment.'
'We couldn't connect to Claude. Please check your network connection and try again.'
'A previous response is still running in this conversation. It will appear shortly — try again once it finishes.'
'Unfortunately, Claude is not available to new users right now. We're working hard to expand our availability soon.'
Claude stops mid-response and never finishes generating
Claude Code hangs, errors, or won't connect in your terminal/IDE
Claude mobile app (iOS/Android) crashes on launch or mid-chat
Anthropic defines 529 as overloaded_error: the API is temporarily overloaded. That is different from a malformed prompt, an expired login, or a problem with your browser. Check the affected component on status.anthropic.com. For API clients, retry with exponential backoff and jitter instead of sending rapid repeat requests.
A 500 api_error is an unexpected error inside Anthropic's API, while 504 is a gateway timeout. Both are normally transient. Preserve the request ID from the response, wait, and retry. If failures continue across simple requests and more than one network or account, treat the problem as an incident rather than rewriting a valid prompt.
Stale site data or an expired web session can cause blank screens, infinite loading, broken controls, login loops, or a repeated error when sending. Test a private window first. If that clean session works, clear only claude.ai site data or sign out and back in before making broader browser changes.
A content-modifying extension, stale site data, or an expired authentication session can leave claude.ai blank, looping, or unable to send. An incognito window is useful because it tests a clean browser state without immediately deleting every cookie. If Claude works there, return to the normal profile and isolate site data or extensions.
Claude usage limits and conversation-length limits are separate. Anthropic currently describes session-based usage that resets every five hours, alongside weekly limits that vary by plan and model. All Claude surfaces associated with a plan can contribute to usage. A long conversation can also exhaust available context even when account usage remains. Check Settings > Usage and read the exact banner before deciding which limit you hit.
Anthropic's documented codes make this diagnosable: 400 means invalid request, 401 authentication, 402 billing, 403 permission, 404 resource not found, 413 request too large, 429 rate limit, and 409 conflict. In Claude Code, a 400 may be cleared by compacting or starting a clean session, but developers should inspect the response body and request ID rather than assuming every 400 is a context problem.
Corporate networks, school networks, and some VPN configurations block claude.ai or Anthropic's API endpoints. Common in enterprise environments that whitelist AI tools individually. Symptoms: requests hang indefinitely rather than returning a clean error.
The Claude iOS/Android apps and the Claude Code CLI update frequently. Running an old version after Anthropic makes backend changes causes crashes, blank responses, sync failures, and avoidable 400s. App stores and package managers don't always auto-update immediately.
When to try: Always first, before changing local settings
Open status.anthropic.com and check the component you are actually using. Claude.ai, the API and Claude Code can be affected differently during a partial incident. If the official page shows a matching incident, stop clearing browser data and wait for the service update. If it is green, test a tiny request in a new chat or session. A tiny request working while one long conversation fails points to conversation state or length, not a platform-wide outage.
When to try: Whenever a numeric error code is visible
For API and Claude Code errors, save the response body and request ID. Treat 400 as an invalid request, 401 as bad or missing authentication, 402 as a billing problem, 403 as insufficient permission, 404 as a missing resource or model, 413 as an oversized request, 429 as rate limiting, 500 as an internal API error, 504 as a timeout, and 529 as overload. The code narrows the investigation far more reliably than the phrase 'Claude is broken.'
When to try: First local fix, 30 seconds, no data loss
Press Ctrl+Shift+R (Windows/Linux) or Cmd+Shift+R (Mac) in your browser to bypass the cache and force a full reload of claude.ai. This resolves most 'blank screen after update' and 'error sending message' issues without needing to clear all cookies. Try this before the more disruptive cache-clear step.
When to try: If hard refresh didn't resolve the issue
In Chrome: Settings → Privacy and security → Third-party cookies → See all site data and permissions → search for 'claude' and 'anthropic' → Delete all. In Firefox: Address bar, click the lock icon → Clear cookies and site data. In Safari: Settings → Privacy → Manage Website Data → search 'anthropic' → Remove. Sign back in after clearing.
When to try: To isolate browser-profile and extension conflicts
Open a private window, sign in to claude.ai and send a short test prompt. If it works there but not in your normal profile, the difference is local browser state: an extension, stored site data, proxy setting or profile policy. Return to the normal profile and disable content-modifying extensions one at a time, then clear claude.ai site data if needed.
When to try: If Claude loads but will not continue a conversation
Open Settings > Usage and read any reset information Anthropic shows for your plan. A session allowance currently resets every five hours, and some plans also have weekly limits. If usage remains, move the task to a new conversation and provide a compact handoff summary; one oversized or damaged thread can fail while a fresh thread works. Paid users who have reached included usage can review Anthropic's usage-credit option, where available, rather than assuming the service is down.
When to try: For Claude Code, API, or integration failures
Start with the exact code. For a 400 in Claude Code, run /compact if the session is long, test a fresh session, and update Claude Code using the installation method you originally chose. For 401, re-authenticate or verify the API key source. For 402, inspect billing or credits. For 403, verify workspace and model permissions. For 413, reduce the request or uploaded material. For 429, 500, 504 and 529, use bounded exponential backoff with jitter. Anthropic's official SDKs retry transient connection errors, rate limits and 5xx responses twice by default and honor retry-after headers, but production code should still define a maximum retry budget and log request IDs.
When to try: For Claude mobile app issues specifically
For iOS: App Store → your account icon → scroll to Claude → Update. For Android: Play Store → My apps → Claude → Update. If already up to date and still broken, delete the app and reinstall it fresh. This clears all cached data and corrupted local state. You won't lose conversation history since it's stored server-side.
When to try: If the app loads but requests always hang or time out
Switch from WiFi to mobile data (or vice versa) to test whether your network is the issue. Disable VPN if active. If on a corporate or school network, try from a personal device on mobile data. Claude is sometimes blocked by network-level content filters before the error reaches your browser.
When to try: If earlier isolation steps have not identified the cause
Try another current browser or device. If the same account works elsewhere, the problem is local to the original browser profile, operating system, app installation or network path. A new browser profile with no extensions is a stronger diagnostic than repeatedly reinstalling the same app because it gives you a clean state to compare.
Bookmark status.anthropic.com, it's the fastest way to confirm outages
Keep the Claude mobile app updated to the latest version to avoid compatibility issues
Use a separate browser profile for Claude if you rely heavily on extensions in your main profile
If you use Claude for critical work, review Settings > Usage and plan around the displayed session and weekly limits
For API work, log the status code, request ID, model, request size, retry count and timestamp without logging sensitive prompt content
Use bounded exponential backoff with jitter for 429, 500, 504 and 529 responses; do not create an infinite retry loop
Claude Projects sync conversation history server-side, you won't lose context if you need to clear local cache
Contact Anthropic support through support.anthropic.com when the official status page is operational but the same account problem survives a clean browser, another network and a fresh conversation; when access, billing or usage appears inconsistent with your plan; or when a repeatable API request keeps failing after you have validated the documented requirements. Include the timestamp, surface, status code, request ID, account or workspace context and the troubleshooting steps already taken. Do not send an API key, password or confidential prompt content.
Incomplete responses happen for a few reasons: (1) Network interruption during streaming, (2) Response hit Claude's output length limit (very long generations sometimes cut off), (3) Content policy filter triggered partway through, (4) Server-side timeout on Anthropic's end during high load. Fix: refresh the page and rephrase the request into shorter steps. If asking for something very long, break it into parts.
The authoritative source is status.anthropic.com. Check whether the affected component matches what you use: Claude.ai, the API and Claude Code can have different states. A third-party outage monitor can provide a useful second signal, but user reports do not prove the cause. If the official status is green, test a fresh, short conversation and another network before treating the issue as a platform outage.
Anthropic currently uses session-based limits that reset every five hours and also applies weekly limits to some plans and models. The exact amount varies with plan, model, message length, attached files and feature use, so a fixed message count is misleading. Usage across Claude surfaces associated with the plan can contribute to the allowance. Check Settings > Usage and the banner in Claude for the reset time and the limit you reached. Eligible paid users can also review usage credits for continuing beyond included usage.
Yes. Claude has native apps for iOS (App Store) and Android (Play Store) as of 2024. The mobile apps support full conversation history, Projects, and file uploads. Common mobile issues: app crashes after updates (fix: reinstall), slow responses on cellular (fix: switch to WiFi), and login loops (fix: sign out fully and back in).
The web interface (claude.ai) and API use the same underlying models but have slightly different system prompts and context configurations. Web Claude has a 'helpful assistant' system prompt built in; raw API calls start without one. Output format also differs since web is optimized for markdown rendering. This is expected behavior, not a bug.
Do not diagnose either product by analogy. Each platform has its own status page, plan limits, request rules and client behavior. For Claude, use Anthropic's status page and documented API error codes; for ChatGPT, use OpenAI's status and documentation. The visible wording can change, so the most reliable evidence is the affected surface, exact code, request ID and account usage display.
No. Claude is a cloud-only service with no offline mode. All inference runs on Anthropic's servers. There's no local version of Claude available to end users (unlike some open-source models you can run locally). You need an active internet connection for every message.
Slow output can come from service load, the selected model, extended reasoning, tool calls, a large context, file processing or your network. Check status.anthropic.com, then compare a short prompt in a new conversation. If the short prompt is responsive but the original thread remains slow, the conversation or task is the likely cause. If all requests return 529 or time out across clients, follow the outage and retry guidance rather than repeatedly resending.
Start with status.anthropic.com and match the affected component to your surface. A third-party report spike is supporting evidence, not proof. If the official component is healthy, test a short prompt in a fresh conversation, a private browser window and another network. Those comparisons reveal whether the failure follows one conversation, one browser profile, one network or the account itself.
Open status.anthropic.com and inspect the relevant component and incident history. Compare the incident start time with your first failure. If you operate an API integration, correlate that window with status codes and request IDs in your logs. Community reports can confirm that other users are affected, but the official component status is the source to use for operational decisions.
Claude Code runs on Anthropic's API, so when Claude Code can't connect or errors out, it usually reflects the same Anthropic incident shown on status.anthropic.com (check the API component specifically). If the status page is green but Claude Code still fails, check your API key/credits, your network or proxy, and that your CLI/IDE extension is updated. A wide Anthropic outage takes Claude.ai, the API, and Claude Code down together.
The selected model may be unavailable to your plan, workspace or region, temporarily affected by an incident, or no longer available through that surface. Check your model selector and plan access, then inspect the relevant status component. For API calls, verify the exact model identifier in current Anthropic documentation instead of substituting a guessed version.
These are transient errors. 'Claude couldn't finish this response — try again in a moment' usually means a server-side timeout or interruption during generation under load; just resend, ideally in smaller steps. 'We couldn't connect to Claude — check your network' points to a connection problem on your side (network, VPN, firewall) or a brief outage; switch network, disable VPN, and retry. If they persist for everyone, check the status page.
The message 'Unfortunately, Claude is not available to new users right now. We're working hard to expand our availability soon' appears during periods when Anthropic temporarily limits new sign-ups due to capacity. It's not an error you can fix — existing users are usually unaffected. Try again later, or sign in if you already have an account. It typically eases as Anthropic scales capacity.
Anthropic documents 529 as overloaded_error: the API is temporarily overloaded. Wait and retry with exponential backoff and random jitter. Respect any retry-after header, cap the number of attempts, and avoid synchronized retries from many workers. The official Anthropic SDKs retry many transient failures twice by default, but applications still need observability and a graceful fallback when the retry budget is exhausted.
A 500 is 'api_error' — an internal server error inside Anthropic's infrastructure, distinct from a 529 overload. It's typically transient: retry the request, ideally with a short backoff, and it usually succeeds. If 500s persist across requests and accounts, it's an incident — check status.anthropic.com. Because it originates server-side, you can't fix a 500 by changing your prompt, key, or network.
A 400 is invalid_request_error, so inspect the accompanying message rather than treating it as an outage. If a long Claude Code session is the trigger, run /compact or start a fresh session. Update Claude Code through the same installation channel you used originally. If the error is repeatable in a short clean session, capture the request ID and exact wording for support. A 413 specifically means the request is too large; a 529 means overload and needs retry behavior instead.
That web-app message does not identify one cause by itself. The send may have failed because of a brief connection interruption, expired session, browser extension, service incident, usage limit or conversation state. Save your unsent text, try once in a new conversation, check the official status page, then use a private window and another network to isolate the cause. Avoid rapid resubmission if the service is returning overload errors.
Use evidence from the affected surface. Check status.anthropic.com, note whether other conversations work, and test the same account in a clean browser or on another network. A confirmed incident or repeated 5xx/529 responses across clients indicates an Anthropic-side problem. A failure isolated to one conversation, browser profile or device calls for local troubleshooting instead.
Claude Code runs on Anthropic's API, so 'degraded performance' there usually mirrors an API-side incident — check the API component on status.anthropic.com, which posts degraded-performance notices separately from full outages. During these windows you may see slower responses, more 529 overload errors, or retries. It's temporary and not something you fix locally; if you need to keep moving, lower the effort/model for routine work. If the status page is green but only your sessions are slow, update Claude Code and check your network/proxy.
Anthropic maintains the Claude Code repository and issue tracker on GitHub, and it's a useful place to confirm whether others are hitting the same error today and whether a fix is in progress. Search the issues for your exact error string (e.g. 'API Error: 400' or '529'). For service status it's still faster to check status.anthropic.com, but GitHub issues are the best signal for CLI-specific bugs, regressions after an update, and workarounds posted by the maintainers and community.
If you were using Claude to build a website or app and you're blocked by an outage, you don't have to stall. Lovable is a separate AI app builder that turns a plain-English prompt into a deployed full-stack app (with a database, auth, and a live URL) — a practical way to keep shipping while you wait. It has a free plan, so you can pick up where you left off. Note that during a very wide Anthropic outage some AI tools can be affected generally; Lovable is its own platform and is often available, and it's a strong alternative builder regardless.
Product behavior and limits can change. These primary sources were used to verify this guide.