LangChain 1.0 and LangGraph 1.0 Are Out — Finally a Stable API You Can Pin in Production

LangChain 1.0 and LangGraph 1.0 shipped July 22 — after four years of 0.x breaking changes, both frameworks commit to no breaking changes between minor releases.

After a 0.x era where a routine pip install -U langchain could silently break production agents, both frameworks crossed 1.0 on July 22. Here's what changed.

LangChain 1.0:

• Middleware layer for agent loops — inject logging, tracing, rate-limiting, and retry logic without subclassing chain internals
• Fully decoupled provider packages — langchain-anthropic, langchain-openai, langchain-google-genai are independent; pin them separately from langchain-core
• Stability contract — no breaking changes between 1.x minor releases; breaking changes require a major bump and a 6-month deprecation window

LangGraph 1.0:

• Typed state (TypedDict or Pydantic) is now required, not optional
• interrupt and Command for human-in-the-loop flows are stable API
• Subgraph composition stable — import compiled graphs as packages

Upgrade:

pip install -U langchain-core langchain-anthropic langgraph

Check startup logs for deprecation warnings — v1.0 removed several 0.x patterns. The migration guide at the announcement covers the most common breaks.

Why it matters: This is the moment LangChain becomes suitable for production code you actually want to maintain. Pin langchain-core>=1.0,<2.0 and upgrade on your own schedule.

Sources: LangChain and LangGraph 1.0 — LangChain Blog · LangChain Changelog