google.genai.errors.ClientError: Input token count exceeds maximum limit

[in0vik]

The application is encountering a google.genai.errors.ClientError: 400 INVALID_ARGUMENT when calling the Google Gen AI API. The specific error message indicates that the input prompt's token count exceeds the model's maximum allowed limit (for big codebase).

Error Message:

google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'The input token count (1094897) exceeds the maximum number of tokens allowed (1000000).', 'status': 'INVALID_ARGUMENT'}}

Traceback:

Traceback (most recent call last):
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge-2/main.py", line 81, in main
    tutorial_flow.run(shared)
    ~~~~~~~~~~~~~~~~~^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 50, in _run
    o=self._orch(shared); return self.post(shared,p,o)
     ~~~~~~~~~~^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 48, in _orch
    last_action=curr._run(shared); curr=copy.copy(self.get_next_node(curr,last_action))
                ~~~~~~~~~^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 13, in _run
    e=self._exec(p); return self.post(shared,p,e)
     ~~~~~~~~~~^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 33, in _exec
    if self.cur_retry==self.max_retries-1: return self.exec_fallback(prep_res,e)
                                                  ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 28, in exec_fallback
    raise exc
    ^^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/pocketflow/__init__.py", line 31, in _exec
    try: return self.exec(prep_res)
                ~~~~~~~~~^^^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge-2/nodes.py", line 149, in exec
    response = call_llm(prompt)
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge-2/utils/call_llm.py", line 56, in call_llm
    response = client.models.generate_content(
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/google/genai/models.py", line 5019, in generate_content
    response = self._generate_content(
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/google/genai/models.py", line 3995, in _generate_content
    response_dict = self._api_client.request(
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/google/genai/_api_client.py", line 726, in request
    response = self._request(http_request, stream=False)
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/google/genai/_api_client.py", line 655, in _request
    errors.APIError.raise_for_response(response)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^
  File "/Users/ilyanovik/Documents/Projects/Tutorial-Codebase-Knowledge/.venv/lib/python3.13/site-packages/google/genai/errors.py", line 101, in raise_for_response
    raise ClientError(status_code, response_json, response)
google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'The input token count (1094897) exceeds the maximum number of tokens allowed (1000000).', 'status': 'INVALID_ARGUMENT'}}

[zachary62]

Google Gemini 2.5 pro has 1M token limit. For large repo, try to cut some large files but non-important files?

Also, according to Google, 2M token limit coming soon...

[in0vik]

Thanks but this doesn't work for me :)

I've been working with a massive codebase (around 2.5M tokens) and kept running into Gemini's 1M token limit.
Ideas:

• Split the codebase into logical components to process separately (kind of chunks)
• Filter out non-essential files (tests, docs, generated code, etc)
• Kind of RAG
• Build a pipeline that first understands the high-level structure before diving into details