Claude AI Error Messages: What Each One Means and the Fix

Most Claude AI errors fall into three groups, and each group has a different fix. You hit a limit (wait for the reset time in the message, or turn on usage credits). Claude is busy or broken (a 529, a 500, or a "capacity constraints" message: wait a few minutes and check status.claude.com). Something on your side is wrong (a bad API key, a request that's too big, a VPN, a stale login). Read the exact wording of the error first. It usually tells you which group you're in.

This guide covers the claude.ai web, desktop, and mobile apps, the Claude API, and Claude Code. Every error name below is quoted from Anthropic's own documentation.

Claude AI Error Quick Reference

Why Is Claude Not Working? Errors in the claude.ai App

If you use Claude in a browser, the desktop app, or on your phone, you'll mostly see plain-English messages instead of status codes. Anthropic's support article on troubleshooting Claude error messages lists five kinds.

Usage limit reached

Anthropic's help center says your plan has a five-hour session limit and a weekly limit, and some plans carry an extra weekly limit for specific models. As you get close, you'll see "Approaching session limit" or "Approaching weekly limit," sometimes with a percentage.

When you hit the wall without usage credits turned on, the message reads either "5-hour limit reached - resets [time]" or "You've reached your weekly limit. It resets [day] at [time]."

The fix: wait for the time in the message. On a paid plan you can turn on usage credits in your Usage settings to keep working on pay-as-you-go billing, and the message changes to "Now using credits."

Anthropic changes the allowance for each plan, so trust its page on how usage and length limits work over any number you read elsewhere. Usage from claude.ai, Claude Code, and the desktop app counts toward the same limit, so a heavy coding session can be why chat stops working. Our guide to Claude Code usage limits explains how the windows overlap.

"Your message will exceed the length limit for this chat"

Your input is too long. Attach fewer or smaller files, split the content into chunks, or start a new conversation. On paid plans with code execution turned on, Claude summarizes long chats on its own, so this mostly comes from a very large first message. If you're not sure whether the problem is file size or token count, check the Claude upload limit for the file caps and the Claude context window for how much text each model can hold.

"Due to unexpected capacity constraints, Claude is unable to respond"

This is Claude being busy, not broken. Anthropic says capacity constraints are not outages, and it doesn't list them on the status page, so a green status page doesn't mean the message is wrong. Try again in a few minutes.

Login errors

For "There was an error logging you in," Anthropic recommends turning off any VPN, disabling browser extensions, and clearing your cache and cookies.

Is Claude down? How to check for outages

A real outage means Claude is unavailable or badly degraded for most users. Anthropic posts confirmed incidents at status.claude.com (the old status.anthropic.com address redirects there). If an incident is listed, there's nothing to fix on your end.

Claude API Errors: 400, 401, 403, 413, 429, 500 and 529

Developers calling the API get HTTP status codes with a named error type. Anthropic's Claude API errors reference is the source for everything in this section.

Every API error comes back as JSON with a top-level error object holding a type and a message, plus a request_id:

{
  "type": "error",
  "error": {
    "type": "not_found_error",
    "message": "The requested resource could not be found."
  },
  "request_id": "req_011CSHoEeqs5C35K2UUqR7Fy"
}

Save the request_id. Anthropic asks for it when you contact support about a specific failed call.

Client errors you need to fix

These are about your request, your key, or your account. Retrying the same request will fail the same way.

• 400 invalid_request_error: the format or content of the request is wrong. It also comes back when your organization or workspace reaches a spend limit you set yourself. In 2026 the common causes are model-specific, such as a prefilled assistant message or a thinking setting the model doesn't accept. The message names the exact problem.
• 401 authentication_error: your API key is malformed, revoked, or expired. Create a new key in the Claude Console.
• 403 permission_error: the key is valid but can't use that resource. Check your organization's access and workspace settings.
• 413 request_too_large: the request is over the maximum byte size. The Messages, Batch, and Files APIs each have their own cap. Upload large documents through the Files API instead of inlining them.

429 rate_limit_error

A 429 means your organization hit a rate limit, reached its usage tier's monthly spend cap, or hit a spend limit on the Claude Code workspace. There's one detail that saves a lot of confused retrying: a tier spend-cap 429 has no retry-after header and keeps failing until access resumes. If your 429s come with a retry-after header, back off and retry. If they don't, you've hit a cap, and retrying won't help.

The rate limits documentation explains how tiers and per-workspace spend limits work. A sudden spike in usage can also trigger 429s through acceleration limits, so ramp traffic up gradually.

Server errors: 500, 504 and 529

These are on Anthropic's side, so the fix is patience and good retry logic.

• 500 api_error: an unexpected error inside Anthropic's systems. Retry with exponential backoff, and contact support with the request ID if it keeps happening.
• 504 timeout_error: the request timed out while processing. For long generations, use the streaming Messages API or the Message Batches API instead of one long non-streaming call.
• 529 overloaded_error: the API is temporarily overloaded, usually from high traffic across all users.

The official SDKs already retry connection errors, 429s, and 5xx errors twice by default with backoff, so if you see one through an SDK, those retries have run. Raise the count with max_retries (maxRetries in TypeScript). With streaming, an error can arrive as an event after a 200 response, so watch the stream too.

Claude Code Errors and How to Fix Them

Claude Code wraps the API errors in its own messages and adds a few more. Anthropic's Claude Code error reference lists them all; these are the common ones. Claude Code retries transient failures up to 10 times before it shows you an error, so the automatic retries are already used up by the time you see one.

"API Error: 500" and "Repeated 529 Overloaded errors"

Claude Code shows these as API Error: 500 Internal server error and API Error: Repeated 529 Overloaded errors. Anthropic's docs say a 529 is not your usage limit and doesn't count against your quota.

The fix: check status.claude.com, wait a minute, then type try again (your original message is still in the conversation). For 529s, capacity is tracked per model, so switching with /model often gets you working right away.

"Request rejected (429)" vs "You've hit your session limit"

These look alike but aren't the same problem.

• You've hit your session limit · resets 3:45pm (or weekly, Opus, or Sonnet limit) is your subscription allowance. Wait for the reset, or run /usage. An Opus or Sonnet limit can be sidestepped with /model.
• API Error: Request rejected (429) is a rate limit on your API key or cloud project. Run /status: a stray ANTHROPIC_API_KEY in your shell can route requests through a low-tier key instead of your subscription.
• API Error: Server is temporarily limiting requests (not your usage limit) is a short server-side throttle. Wait briefly and retry.

Login and authentication errors

• Not logged in · Please run /login: no valid credential. Run /login.
• Invalid API key · Fix external API key: the key in ANTHROPIC_API_KEY was rejected. Run env | grep ANTHROPIC to find a stale key, often loaded from a project .env file, or unset it and use /login instead.
• OAuth token revoked · Please run /login: your saved login is no longer valid. Sign in again.
• API Error: 403 Request not allowed: the installation and login troubleshooting page says to confirm your Pro or Max subscription is active, or that your Console account has the Claude Code or Developer role. A corporate proxy can also cause this.

Context and size errors

• Context limit reached · /compact or /clear to continue: the conversation plus files is bigger than the model's context window. Run /compact to summarize, or /clear to start fresh.
• Request too large (max 32MB): accumulated images or pasted files pushed the request over the limit. Run /compact, and reference large files by path instead of pasting them.
• There's an issue with the selected model: the model name isn't recognized or your account can't use it. Run /model and pick from the list.

If Claude Code won't start at all, run claude doctor from your shell. Many odd errors disappear after an update, and how to update Claude Code covers why auto-update fails on some install methods.

A Fix Order for Most Claude Issues

1. Check status.claude.com. If there's an incident, wait.
2. Retry after a few minutes. That clears capacity messages, 500s, and 529s.
3. Check your credential. /status in Claude Code, a current key on the API, no VPN in the browser.
4. Shrink the request. Fewer files, /compact, or a new chat.
5. Switch models. Capacity is tracked per model.
6. Report it with the request ID, or /feedback in Claude Code, if a 500 keeps happening.

Running Claude Without Fighting the Setup

Many Claude Code errors come from the setup around it: environment variables, API keys, and proxies. If that layer is where you keep getting stuck, Taku mirrors a working AI setup into a desktop workspace and runs it, so you can use a power user's configuration without rebuilding their environment. Taku is in Beta, and the Mac app is available now.

Key Takeaways

• A reset time means a limit, a status code means an API issue, and "capacity" means try again later.
• Capacity messages and 529s aren't outages and won't appear on the status page.
• A 429 without a retry-after header is a spend cap; retrying won't help.
• 400, 401, 403, and 413 need a change on your side.
• In Claude Code, /status, /model, /compact, and /login fix most errors.

FAQ

Why is Claude not working right now?

Check status.claude.com first. If there's no incident, the most likely causes are a usage limit (the message shows a reset time), a capacity constraint (try again in a few minutes), or a local problem such as a VPN or browser extension.

What's the difference between a 429 and a 529 error?

A 429 rate_limit_error is about your organization: you hit a rate limit or spend cap. A 529 overloaded_error is about Anthropic: the API is at capacity for everyone. A 429 can need a tier or spend-limit change. A 529 just needs time.

Does a usage limit on claude.ai affect Claude Code?

Yes. Anthropic says usage across claude.ai, Claude Code, and the desktop app counts toward the same limit, so heavy use in one can block the others until the reset.

Where do I report a Claude error that keeps happening?

For the API, contact Anthropic support and include the request_id from the error response. In Claude Code, run /feedback, which sends your request details to Anthropic.