Error messages

API Error: Request timed out.

API Error: Request timed out.

This is Claude Code saying a request to Anthropic's API ran out of time before an answer came back. It isn't a limit on your account and nothing is wrong with your prompt. The full line looks like this:

API Error: Request timed out.

"Request timed out." is the stock message of the Anthropic SDK that Claude Code bundles, and it's still there in version 2.1.280, with Claude Code's API Error prefix in front. 273 issues in the anthropics/claude-code repo mention "Request timed out" as of September 23, 2026. We read the binary and Anthropic's error reference to sort out which of the several timeouts you've hit.

What the error reference says causes it

Anthropic's error reference puts it plainly: "The API didn't respond before the connection deadline." It gives two usual causes, periods of high load and a model generating a very large response, and a default request timeout of 10 minutes. That second cause is easy to miss. A long refactor that asks for one huge answer can run into the deadline on a perfectly healthy connection.

Three messages, three different clocks

The version with a longer tail, Request timed out. Check your internet connection and proxy settings, comes from a different place. In the 2.1.280 code it's what Claude Code prints when the operating system reports ETIMEDOUT, meaning the network connection itself timed out. The error reference files that one under network problems, next to Connection dropped (ECONNRESET).

Then there's No response from API. Before v2.1.242, Claude Code sat through the whole 10 minutes when the API never sent back response headers. Now a shorter first-byte deadline catches that case, Claude Code tries once more, and if the retry also gets nothing it ends the turn and tells you how long each attempt waited.

What you'll see before it gives up

Claude Code retries timeouts that happen before any of the reply has streamed, up to 10 times by default. If no data arrives for 20 seconds, the spinner shows Waiting for API response · will retry in … · check your network while it waits. The request hasn't failed at that point. In a June 2025 report on version 1.0.35 the person saw the retries count all the way to 10 of 10, then got the timeout after a file read, switched to Sonnet, and got it again. Changing model didn't help.

What to do

Send the message again first. For long jobs, ask for the work in smaller pieces so no single response has to run for minutes. If you're behind a corporate proxy or a gateway that holds the whole response until it's finished, raise API_TIMEOUT_MS (it's in milliseconds, and the default is 600000). On v2.1.242 or later, CLAUDE_STREAM_FIRST_BYTE_TIMEOUT_MS sets the shorter wait for the first byte.

When it happens on every request, Boris Cherny from Anthropic's Claude Code team gave three checks on that 2025 thread: look for a gateway or proxy slowing requests down, confirm DNS and your connection work, and make sure you can reach the API with curl by hand. The error reference's version of that last one is curl -I https://api.anthropic.com.

A 529 is a separate problem, where the API answered and said it was busy. That's covered on Repeated 529 Overloaded errors.