ChatAnthropic integration: install, model names, and key handling

ChatAnthropic

pip install -U "langchain[anthropic]"

from langchain_anthropic import ChatAnthropic

model = ChatAnthropic(model="claude-sonnet-4-6")

Or provider-agnostic:

from langchain.chat_models import init_chat_model

model = init_chat_model("claude-sonnet-4-6")

Rules

• The key comes from ANTHROPICAPIKEY in the environment. If auth fails, check the variable is actually set in the process environment (dotenv loading order bites people) before assuming the key is bad.
• Claude is one of the providers that supports standard content blocks (message.content_blocks), so prefer that accessor when you want reasoning or tool-use content in a provider-agnostic shape.
• For long agentic runs, bump max_retries (default 6) and set a checkpointer so a transient failure does not lose the conversation.

Verified against the official LangChain docs on 2026-09-26. Models: https://docs.langchain.com/oss/python/langchain/models