OpenAI Integration for LangChain

1.1.12 · active · verified Sat Mar 28

An integration package connecting OpenAI and LangChain, providing classes for chat models (`ChatOpenAI`) and embeddings (`OpenAIEmbeddings`). It is actively maintained with frequent releases, typically minor versions every few months and patch releases more frequently, as part of the broader LangChain ecosystem.

Warnings

breaking LangChain v1 introduced a new package structure. Most OpenAI-related classes moved from the monolithic `langchain` package (e.g., `langchain.chat_models.ChatOpenAI`) to the dedicated `langchain-openai` partner package (e.g., `langchain_openai.ChatOpenAI`). Code using old import paths will break.
Fix: Update import statements from `from langchain...` to `from langchain_openai...` for OpenAI-specific components like `ChatOpenAI` and `OpenAIEmbeddings`.
gotcha Compatibility with `langchain-core` is crucial. The `langchain-openai` package frequently bumps its minimum required `langchain-core` version. Running `langchain-openai` with an outdated `langchain-core` can lead to `AttributeError`s or unexpected behavior due to schema drifts or missing fields. [cite: 1.1.12 release notes]
Fix: Always ensure `langchain-core` is up-to-date by running `pip install -U langchain-openai langchain-core`.
breaking For Azure OpenAI, `ChatOpenAI` now supports the v1 API directly using `base_url` and `api_key` parameters since `langchain-openai>=1.0.1`. The legacy `AzureChatOpenAI` class is still available but `ChatOpenAI` is the recommended unified approach for the v1 API.
Fix: For new Azure OpenAI integrations, use `langchain_openai.ChatOpenAI` with `base_url` set to your Azure endpoint (e.g., `https://{your-resource-name}.openai.azure.com/openai/v1/`) and provide the `api_key`. Migrate existing `AzureChatOpenAI` usage if a unified interface is desired.
gotcha When using `OpenAIEmbeddings` with non-OpenAI compatible APIs (e.g., OpenRouter, Ollama, vLLM via `base_url`), the default `check_embedding_ctx_length` might cause errors.
Fix: Set `check_embedding_ctx_length=False` in the `OpenAIEmbeddings` constructor (e.g., `OpenAIEmbeddings(model="your-model", base_url="https://your-api-endpoint.com", check_embedding_ctx_length=False)`).
deprecated When interacting with the OpenAI Responses API, `langchain-openai` now defaults to storing response items in message content. To restore the previous (v0) behavior, users need to explicitly opt-in.
Fix: Set the `LC_OUTPUT_VERSION` environment variable to `v0` or specify `output_version="v0"` when instantiating `ChatOpenAI` (e.g., `ChatOpenAI(model="gpt-4o-mini", output_version="v0")`).

Install

pip install langchain-openai Install package

Imports

ChatOpenAI
```
from langchain_openai import ChatOpenAI
```
As of LangChain v1, `ChatOpenAI` is part of the `langchain-openai` partner package, not `langchain.chat_models` (which is for older v0 usage).
OpenAIEmbeddings
```
from langchain_openai import OpenAIEmbeddings
```
As of LangChain v1, `OpenAIEmbeddings` is part of the `langchain-openai` partner package, not `langchain.embeddings` (which is for older v0 usage).

Quickstart

This quickstart demonstrates how to initialize `ChatOpenAI` and send a basic message. It assumes the `OPENAI_API_KEY` environment variable is set.

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

# Set your OpenAI API key as an environment variable (recommended)
# Or pass it directly: ChatOpenAI(openai_api_key="your_api_key")

# Ensure OPENAI_API_KEY is set in your environment
if not os.environ.get("OPENAI_API_KEY"): 
    os.environ["OPENAI_API_KEY"] = os.environ.get("OPENAI_API_KEY", "") # Replace with actual key or prompt user
    # For interactive use, you might use: 
    # import getpass 
    # os.environ["OPENAI_API_KEY"] = getpass.getpass("Enter your OpenAI API key: ")

# Instantiate the ChatOpenAI model
chat_model = ChatOpenAI(model="gpt-4o-mini", temperature=0.7)

# Invoke the model
response = chat_model.invoke([HumanMessage(content="Tell me a short story about a brave knight.")])

print(response.content)

view raw JSON →