LangChain reached its first stable release on October 22, 2025, and it is more of a reset than an incremental bump. Both the Python and JavaScript packages hit 1.0 together, the sprawling collection of chains and helpers was trimmed down, and the framework now stands on a single agent abstraction (create_agent) running on the LangGraph runtime. A follow up, LangChain 1.1, landed on December 2, 2025. If you build LLM applications, this changes the recommended way to assemble agents and the contract you get back from models.

What happened

LangChain 1.0 made a clean break from the old “chains everywhere” design and committed to a stable surface, with no breaking changes planned until a future 2.0. The headline pieces:

  • create_agent is the new standard entry point for building agents. It replaces both the legacy AgentExecutor and the langgraph.prebuilt.create_react_agent shortcut. You give it a model, tools, and a system prompt, and it runs the basic agent loop: call the model, let it choose tools, finish when no more tools are called.
  • Agent middleware gives you hooks at every step of that loop instead of subclassing. LangChain ships built in middleware for human in the loop approval, message history summarization, and PII redaction, and you can write your own.
  • Standard content blocks arrived as a new content_blocks property on messages. It exposes modern model features (reasoning traces, citations, tool calls) through one provider agnostic shape, supported across langchain-anthropic, langchain-aws, langchain-openai, langchain-google-genai, and langchain-ollama.
  • Legacy functionality moved to a separate langchain-classic package. That includes the old chains, retrievers such as MultiQueryRetriever, the indexing API, the hub module, and langchain-community exports. The core langchain namespace is now focused on agent building blocks.
  • LangGraph 1.0 became the foundational runtime underneath, providing durable state, resumable execution, and human in the loop patterns.

LangChain 1.1, on December 2, 2025, added model profiles. Chat models now expose a .profile attribute describing their capabilities (structured output, function calling, and similar), sourced from the open models.dev catalog. The summarization middleware and native structured output strategy use these profiles to make smarter decisions, and create_agent gained support for SystemMessage instances, built in model retry with backoff, and OpenAI moderation as middleware.

Why it matters for builders

The practical shift is that there is now one blessed way to build an agent rather than a dozen overlapping chains. If you have been choosing between AgentExecutor, create_react_agent, and hand rolled LangGraph, the answer is now create_agent plus middleware. Cross cutting concerns like guardrails, summarization, and approval steps become composable middleware instead of bespoke wrapping code.

Standard content blocks reduce provider lock in at the message layer: code that reads reasoning or citations no longer has to special case each provider’s response format. And because structured output generation is folded into the main agent loop, you avoid the extra model call that older patterns needed, which trims latency and cost.

The trade off is migration work. Anything you relied on from the old chains or retrievers now lives in langchain-classic, so upgrading is not a no op.

What to do

  • New projects: start on create_agent and add middleware for guardrails, summarization, and any human in the loop steps.
  • Existing projects: pin your current version, read the official migration guide, and budget time to either move imports to langchain-classic or refactor onto create_agent.
  • Multi provider apps: adopt content_blocks so reasoning and citation handling is uniform, and lean on model profiles (1.1) to detect capabilities instead of hard coding them.

Sources

  1. LangChain blog. “LangChain and LangGraph Agent Frameworks Reach v1.0 Milestones.” https://www.langchain.com/blog/langchain-langgraph-1dot0
  2. LangChain changelog. “LangChain 1.0 now generally available.” https://changelog.langchain.com/announcements/langchain-1-0-now-generally-available
  3. LangChain changelog. “LangChain 1.1.” https://changelog.langchain.com/announcements/langchain-1-1
  4. LangChain docs. “What’s new in LangChain v1.” https://docs.langchain.com/oss/python/releases/langchain-v1