# Cannot import model class from 'langchain'
The failing pattern (stale):
```python
from langchain.chat_models import ChatOpenAI # ImportError in v1
```
The current patterns:
```python
from langchain_openai import ChatOpenAI
from langchain_anthropic import ChatAnthropic
```
Or provider-agnostic:
```python
from langchain.chat_models import init_chat_model
model = init_chat_model("openai:gpt-5.5")
```
## Why it happens
- v1 reduced the `langchain` namespace to agents, messages, tools, chat models (the init functions), and embeddings. Model classes moved out to `langchain-openai`, `langchain-anthropic`, and friends.
- Old tutorials, StackOverflow answers, and model weights of knowledge all predate this. Agents copy them verbatim.
## Fix
1. Install the provider package: `pip install -U langchain-openai`.
2. Change the import to the provider package, or switch to `init_chat_model` with a `provider:model` string.
3. If the code genuinely needs the old re-export surface, that is what `langchain-classic` is for, but for model classes the provider import is the right fix.