You are viewing the v1 docs for LangChain, which is currently under active development. Learn more.
1.0 Alpha releases are available for the following packages:
  • langchain
  • langchain-core
  • langchain-anthropic
  • langchain-openai
Broader support will be rolled out during the alpha period.

​
New features

LangChain 1.0 introduces new features:
  • A new .content_blocks property on message objects. This property provides a fully typed view of message content and standardizes modern LLM features across providers, including reasoning, citations, server-side tool calls, and more. There are no breaking changes associated with the new message content. Refer to the message content docs for more info.
  • New prebuilt langgraph chains and agents in langchain. The surface area of the langchain package has been reduced to focus on popular and essential abstractions. A new langchain-legacy package is available for backward compatibility. Refer to the new agents docs and to the release notes for more detail.

​
Breaking changes

​
Deprecations

​
Prebuilt agents

The langchain release focuses on reducing LangChain’s surface area and narrowing in on popular and essential abstractions.

​
ReAct agent migration

create_react_agent has moved from langgraph to langchain with significant enhancements: Enhanced structured output create_react_agent enhances coercion of outputs to structured data types: 
from langchain.agents import create_react_agent
from langchain_core.messages import HumanMessage
from pydantic import BaseModel

class Weather(BaseModel):
    temperature: float
    condition: str

def weather_tool(city: str) -> str:
    """Get the weather for a city."""
    return f"it's sunny and 70 degrees in {city}"

agent = create_react_agent(
    "openai:gpt-4o-mini",
    tools=[weather_tool],
    response_format=Weather
)
result = agent.invoke({"messages": [HumanMessage("What's the weather in SF?")]})
print(repr(result["structured_response"]))
#> Weather(temperature=70.0, condition='sunny')
Structural improvements
  • Main loop integration: Structured output is now generated in the main loop instead of requiring an additional LLM call
  • Tool/output choice: Models can choose between calling tools, generating structured output, or both
  • Cost reduction: Eliminates extra expense from additional LLM calls
Advanced configuration Two strategies for structured output generation:
  1. Artificial tool calling (default for most models)
    • LangChain generates tools matching your response format schema
    • Model calls these tools, LangChain coerces args to desired format
  2. Provider implementations
    • Uses native structured output support when available
    • Configure with ProviderStrategy hint
Prompted output is no longer supported via the response_format argument.

​
Error handling

Structured output errors Control error handling via the handle_errors arg to ToolStrategy:
  • Parsing errors: Model generates data that doesn’t match desired structure
  • Multiple tool calls: Model generates 2+ tool calls for structured output schemas
Tool calling errors Updated error handling for tool failures:
  • Invocation failure: Agent returns artificial ToolMessage asking model to retry (unchanged)
  • Execution failure: Agent now raises ToolException by default instead of retrying (prevents infinite loops)
Configure behavior via handle_tool_errors arg to ToolNode.

​
Breaking changes

Pre-bound models To better support structured output, create_react_agent no longer supports pre-bound models with tools or configuration: 
# No longer supported
model_with_tools = ChatOpenAI().bind_tools([some_tool])
agent = create_react_agent(model_with_tools, tools=[])

# Use instead
agent = create_react_agent("openai:gpt-4o-mini", tools=[some_tool])
Dynamic model functions can return pre-bound models if structured output is not used.
Import changes 
# Before
from langgraph.prebuilts import create_react_agent, ToolNode, AgentState

# After
from langchain.agents import create_react_agent, ToolNode, AgentState

​
Reporting issues

Please report any issues discovered with 1.0 on GitHub using the 'v1' label.

​
See also