[BUG] API Error Report: Frequent JSONDecodeError with deepseek-reasoner Model

[pythonbabaa]

Issue Description

I am running a series of AI agents using langgraph. Each agent sends API call to deepseek-reasoner model. Sometimes, it gets stuck at first agent, sometimes second or third. I have also implemented five retries per agent but all five retries produce same error.

When making API calls to the deepseek-reasoner model, I am consistently receiving a JSONDecodeError: Expecting value error. This occurs even when the API returns an HTTP 200 status code, but the response body appears to be empty or malformed.

Here is a snippet of the code I am using:

from langchain_openai import ChatOpenAI

llm_deepseek = ChatOpenAI(
model='deepseek-reasoner',
base_url="https://api.deepseek.com/",
api_key="<my_api_key>",
temperature=0,
model_kwargs={"response_format": {"type": "text"}}
)
response = llm_deepseek.invoke("formatted_prompt")

Error Details

The full traceback of the error is as follows:

JSONDecodeError: Expecting value: line 1 column 1 (char 0)

Steps Taken

To troubleshoot, I have:

Verified that my API key is valid and has sufficient quota.

Checked the API status page ([status.deepseek.com](http://status.deepseek.com/)) for any ongoing issues.

Implemented retry logic and fallback mechanisms, but the issue persists.

[mowentian]

For API-related performance updates, please visit: https://status.deepseek.com/