Chat models: init_chat_model

init_chat_model builds a chat model from a "provider:model" string. You call it with .invoke(messages) and get back an AIMessage.

Last updated: 27 Sep, 2026 · LangChain 1.4

A hosted model needs an API key. So the graph lessons can run without one, this course uses a small stand-in model that behaves like a real one; the only change to go live is the two lines below.

The init_chat_model call

from langchain.chat_models import init_chat_model

model = init_chat_model("openai:gpt-4.1")             # "provider:model"
model = init_chat_model("gpt-4.1", model_provider="openai", temperature=0)

reply = model.invoke("Why do parrots talk?")         # returns an AIMessage
print(reply.content)

A stand-in model

The graph lessons need a model, but a hosted one needs an API key. Build a small stand-in instead: a function that returns a fixed AIMessage, so the code runs with no key.

from langchain.messages import HumanMessage, AIMessage

def fake_model_invoke(messages):
    # a real model would read the messages and decide what to say
    return AIMessage("Thanks, I can help with that.")   # the stand-in always says this

Calling the stand-in

Call it the way you would call a real model, passing a list of messages, and read the reply's content.

reply = fake_model_invoke([HumanMessage("I was charged twice.")])
print(reply.content)

The stand-in in a run

The stand-in and the call together:

from langchain.messages import HumanMessage, AIMessage

def fake_model_invoke(messages):
    # a stand-in: a real model would read the messages and think
    return AIMessage("Thanks, I can help with that.")

reply = fake_model_invoke([HumanMessage("I was charged twice.")])
print(reply.content)

What init_chat_model returns

• init_chat_model takes a provider and model name and returns a model object.
• .invoke accepts a string, a message list, or role dicts, and returns an AIMessage.
• The stand-in returns a fixed AIMessage, which is enough to build and test a graph; swapping in init_chat_model is the only change to use a real model.

Two ways to name the model

Where the model call fits

• The one line that turns a plain graph into one that calls a real model.
• Swapping providers is a string change, not a rewrite.

Related

• Previous: MessagesState and add_messages
• Next: Tools and bind_tools
• Reference: Chat models