Chat Models & Providers

LangChain provides a universal interface for chat models across providers. Instead of importing provider-specific classes, you can use init_chat_model to instantiate any supported model with a single function call.

Initializing a Chat Model

The init_chat_model function lets you create a chat model by specifying the model name and provider:

from langchain.chat_models import init_chat_model
 
model = init_chat_model("gpt-4o-mini", model_provider="openai")
response = model.invoke("What is LangChain?")
print(response.content)

Supported Providers

init_chat_model supports all major LLM providers out of the box:

Each provider requires its own API key set as an environment variable (OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY).

Key Parameters

You can pass configuration parameters when initializing a model:

model = init_chat_model(
    "gpt-4o-mini",
    model_provider="openai",
    temperature=0.7,
    max_tokens=256,
    timeout=30,
)

Invoking a Model

With a Simple String

The simplest way to call a model is with a plain string:

response = model.invoke("Explain AI agents in one sentence.")
print(response.content)

With a Message List

For more control, pass a list of messages:

from langchain_core.messages import SystemMessage, HumanMessage
 
messages = [
    SystemMessage(content="You are a concise assistant."),
    HumanMessage(content="What is LangChain?"),
]
 
response = model.invoke(messages)
print(response.content)

Switching Providers Seamlessly

The power of init_chat_model is that your application code stays the same regardless of the provider. Only the initialization changes:

openai_model = init_chat_model("gpt-4o-mini", model_provider="openai")
anthropic_model = init_chat_model("claude-sonnet-4-20250514", model_provider="anthropic")
google_model = init_chat_model("gemini-2.0-flash", model_provider="google_genai")
 
question = "What is an AI agent?"
 
for model in [openai_model, anthropic_model, google_model]:
    response = model.invoke(question)
    print(response.content)
    print("---")

This makes it trivial to benchmark models, switch providers, or let users choose their preferred model at runtime.

Configurable Models

You can create a model that defers the provider choice until invocation time:

configurable_model = init_chat_model(configurable_fields="any")
 
result = configurable_model.invoke(
    "Hello",
    config={"configurable": {"model": "gpt-4o-mini", "model_provider": "openai"}},
)
print(result.content)

Key Takeaways

• init_chat_model provides a single entry point for all chat model providers
• Switching between OpenAI, Anthropic, and Google requires changing only the model name and provider string
• Key parameters like temperature, max_tokens, and timeout work consistently across providers
• Models can be invoked with a simple string or a structured list of messages
• Configurable models let you defer provider selection to runtime