Add support for accessing reasoning output from Groq models

[kissycn]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

The Groq API supports a reasoning_format parameter that allows accessing the model's reasoning process separately from its final output. This is particularly useful for verification, explanation, and debugging purposes. While the current langchain_groq integration correctly passes the reasoning_format parameter to the Groq API through model_kwargs, it doesn't provide any way to access this reasoning data in the response.

Current Behavior
When using the native Groq Python SDK, we can access reasoning data like this:

from groq import Groq
client = Groq()
completion = client.chat.completions.create(
    model="deepseek-r1-distill-llama-70b",
    messages=[{"role": "user", "content": "How many r's are in the word strawberry?"}],
    reasoning_format="parsed",
    stream=True
)

for chunk in completion:
    print(chunk.choices[0].delta.reasoning or "", end="")  # Access reasoning data

However, when using the LangChain integration, even though the parameter is passed correctly, there's no way to access this reasoning data:

from langchain_groq import ChatGroq
from langchain_core.messages import HumanMessage

groq = ChatGroq(
    model="deepseek-r1-distill-llama-70b",
    model_kwargs={"reasoning_format": "parsed"}
)
messages = [HumanMessage(content="How many r's are in the word strawberry?")]
response = groq.invoke(messages)
# response only contains content, no reasoning data
print(response)

Motivation

I would like to request that the langchain_groq integration be updated to provide access to the reasoning data, perhaps by:

Including reasoning in the generation metadata
Adding a new property or method to access reasoning data
Supporting a callback for reasoning data in streaming mode

This would make it possible to use Groq's reasoning capabilities within the LangChain

Proposal (If applicable)

No response

[mdrxy]

Added in langchain-groq==0.3.4!

Reasoning effort possibly to come...