Most Claude error messages come from infrastructure, not from anything you did. The common ones, like "tool result could not be submitted" or a 529 overloaded error, mean a connection dropped or Anthropic's servers are under load. Nearly all of them clear with a refresh, a retry, or a short wait. This page explains each major error, why it happens, and the fixes that actually work, as of July 2026.
These errors show up constantly in the reports users leave on our live Claude tracker, so we see which ones spike and when. When an error message suddenly floods in from many users at once, it is almost always an Anthropic-side incident rather than a local problem.
"Tool result could not be submitted. The request may have expired or the connection was interrupted. Refresh the page to continue."
What it means. Claude ran a tool on your behalf (web search, code execution, file analysis) and the result could not be delivered back into the conversation. The request that was waiting for the result expired, or the connection between your browser and Anthropic's servers broke while the tool was still running.
Why it happens. Long-running tool calls are the trigger. The longer a tool runs, the bigger the window for a network hiccup, a sleeping laptop, a flaky connection, or a server-side timeout to kill the in-flight request. It also spikes during Anthropic incidents, which you can confirm on the Anthropic status page.
Fixes to try.
- Refresh the page, exactly as the message says. The conversation usually recovers and you can resend the last prompt.
- If the reply that failed was long, ask Claude to continue from where it stopped rather than regenerating everything.
- Keep the tab active during long tool runs; browsers throttle background tabs and sleeping machines drop connections.
- If refreshing loops back to the same error, start a new conversation. The stuck tool state stays behind in the old one.
"Streaming interrupted. Waiting for the complete message"
What it means. Claude streams its answer token by token. This notice appears when the stream broke mid-message and the app is trying to fetch the finished message in one piece instead.
Why it happens. Streaming needs an uninterrupted connection for the whole response. Wifi handoffs, VPNs, proxies, and corporate networks that buffer streamed responses are the usual culprits, along with plain server load.
Fixes to try.
- Wait. In most cases the complete message arrives within a few seconds.
- If nothing arrives, refresh the page; the finished message is usually there.
- If it recurs on every message, try disabling a VPN or switching networks, since something on the path is breaking the stream.
Claude 529 "Overloaded" errors
What it means. A 529 is Anthropic's capacity signal: overloaded_error means the API had more traffic than it could serve at that moment. It is not a rate limit on your account (that is a 429) and not an error in your request.
Why it happens. Demand spikes. They cluster around US working hours, new model releases, and outages elsewhere. Claude Code and API users see 529s more than chat users because their requests hit the API directly.
Fixes to try.
- Retry after a short wait. 529s usually pass within seconds to minutes.
- In scripts and API integrations, retry with exponential backoff rather than hammering the endpoint.
- If 529s persist for a long stretch, check the status page and check our Claude reviews archive to see whether other users are reporting the same wall.
"Something went wrong and the content wasn't generated"
What it means. A generic generation failure in the Claude apps: the response crashed before anything could be shown. It appears most often with artifacts, image analysis, and very long conversations.
Why it happens. The catch-all nature of the message means several causes share it: a generation that hit an internal error, an oversized conversation, a malformed file, or a transient backend failure.
Fixes to try.
- Retry the same prompt once; transient failures usually do not repeat.
- If it repeats in a long conversation, start a fresh one. Conversation size is the most common repeatable cause.
- If a specific file or image triggers it, re-export or re-save the file and upload again.
How can you tell if the problem is you or Anthropic?
Cross-check in this order: the Anthropic status page for acknowledged incidents, then our live Claude sentiment tracker, where error complaints from real users show up in the vote messages within minutes of a real outage. If both look quiet, the cause is probably local: your network, your browser, or one oversized conversation. Details on how we collect this data are on the about page.
FAQ
Why do I keep getting "tool result could not be submitted"?
If it happens repeatedly in one conversation, that conversation is stuck with a dead tool call; start a new one. If it happens across conversations, it correlates with long tool runs on an unstable connection, or with an ongoing Anthropic incident.
Is a 529 error my fault?
No. A 529 means Anthropic's servers were overloaded when your request arrived. Nothing about your prompt, account, or code caused it, and the only fix on your side is retrying with backoff.
Do these errors mean Claude is getting worse?
Errors and model quality are separate things. Errors are infrastructure failing loudly; quality changes are subtler and harder to pin down. If your concern is the second kind, see our data-backed look at whether Claude is getting dumber.