context_length_exceeded

What the error looks like

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "prompt is too long: 247583 tokens > 200000 maximum"
  }
}

The error is surfaced as invalid_request_error with a message containing "prompt is too long" or "context length".

Context window breakdown

Fix 1: Truncate old messages

Simplest approach — drop oldest turns when approaching the limit:

import anthropic

client = anthropic.Anthropic()
MAX_INPUT_TOKENS = 180_000   # leave headroom for response

def count_tokens(messages, model="claude-sonnet-4-6"):
    result = client.messages.count_tokens(
        model=model,
        messages=messages,
    )
    return result.input_tokens

def trim_history(messages, model="claude-sonnet-4-6"):
    """Remove oldest pairs (user+assistant) until under limit."""
    while count_tokens(messages, model) > MAX_INPUT_TOKENS and len(messages) > 2:
        # Remove the oldest user+assistant pair (keep at least last turn)
        messages = messages[2:]
    return messages

# Usage
messages = [...]   # your conversation history
messages = trim_history(messages)

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=4096,
    messages=messages,
)

Fix 2: Summarize long history

Better for preserving context — summarize early turns with a cheap call:

def summarize_history(messages, keep_last=10):
    """Summarize all but the last N messages."""
    if len(messages) <= keep_last:
        return messages

    to_summarize = messages[:-keep_last]
    recent = messages[-keep_last:]

    # Use haiku for cheap summarization
    summary_resp = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=1024,
        messages=[{
            "role": "user",
            "content": (
                "Summarize this conversation history concisely, "
                "preserving key facts and decisions:\n\n"
                + "\n".join(
                    f"{m['role']}: {m['content']}"
                    for m in to_summarize
                    if isinstance(m.get('content'), str)
                )
            )
        }]
    )

    summary_text = summary_resp.content[0].text
    summary_message = {
        "role": "user",
        "content": f"[Earlier conversation summary: {summary_text}]"
    }

    return [summary_message] + recent

Fix 3: Check token count before sending

# Always count tokens before expensive calls
token_count = client.messages.count_tokens(
    model="claude-sonnet-4-6",
    system="Your system prompt here",
    messages=messages,
)

print(f"Input tokens: {token_count.input_tokens} / 200,000")

if token_count.input_tokens > 195_000:
    messages = trim_history(messages)   # or summarize

Fix 4: Use prompt caching for large static context

If you have a large but unchanging system prompt or document, use prompt caching — it doesn't reduce tokens but makes large contexts 90% cheaper and faster:

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system=[{
        "type": "text",
        "text": very_long_document,
        "cache_control": {"type": "ephemeral"},   # cache this!
    }],
    messages=messages,
)

FAQ