Anthropic — context window exceeded error

Anthropic returns an error because the total input tokens plus requested max_tokens exceed the model context window, requiring truncation or a larger model.

intermediatedebugginganthropicllm

Symptoms

error: input length plus max_tokens exceeds model context window
Long conversations fail but short ones succeed
The error occurs after adding a new message to the history

Possible causes

The conversation history is too long for the model context window
max_tokens is set too high, leaving no room for the input
The system prompt plus messages plus max_tokens exceed the limit

Diagnosis steps

Count the input tokens

echo $PROMPT | anthropic token-count --model claude-3-5-sonnet

Interpretation

Compare the token count against the model context window (200K for Sonnet).

Check max_tokens setting

grep -n "max_tokens" src/llm.ts

Interpretation

Ensure max_tokens plus input tokens does not exceed the window.

Root cause

The total of input tokens and max_tokens exceeds the model maximum context window, so the API rejects the request before processing.

Fix

Truncate or summarize older messages, reduce max_tokens, use a model with a larger context window, or implement sliding window conversation management.

Verify the fix

Retry the request with a shorter input or smaller max_tokens; confirm the model responds successfully.

Symptom walkthrough

Anthropic returns an error because the total input tokens plus requested max_tokens exceed the model context window, requiring truncation or a larger model. The error appears in production but may not surface in local development because the conditions that trigger it depend on the deployment environment or runtime configuration. Understanding the full symptom set helps distinguish this issue from similar errors.

Diagnosis in practice

Follow the diagnosis steps in the frontmatter to isolate the cause. Each step includes a command and an interpretation of the output. The goal is to narrow down whether the issue is in the configuration, the runtime, or the application code. Reproduce the error in a minimal setup before applying a fix.

Root cause

The total of input tokens and max_tokens exceeds the model maximum context window, so the API rejects the request before processing.

Fix

Truncate or summarize older messages, reduce max_tokens, use a model with a larger context window, or implement sliding window conversation management. After applying the fix, monitor the system to confirm the error does not recur under load or in different environments.

Verification

Retry the request with a shorter input or smaller max_tokens; confirm the model responds successfully. If the verification passes, the fix is confirmed; if not, re-examine the diagnosis steps for missed contributing factors.

Diagnostic commands

Run these commands to reproduce and diagnose the issue:

echo $PROMPT | anthropic token-count --model claude-3-5-sonnet
grep -n "max_tokens" src/llm.ts