Changelog

Status: ACTIVE (pulled from docs.langchain.com) Source: https://docs.langchain.com/oss/python/langgraph/changelog-py Timestamp: 2026-05-11

Key release notes for LangGraph and related packages.

langgraph v1.1 (Mar 10, 2026)

Type-safe streaming (version="v2"): Unified StreamPart output with type, ns, and data keys. Each mode has its own TypedDict.
Type-safe invoke (version="v2"): GraphOutput object with .value and .interrupts attributes.
Pydantic and dataclass coercion: Auto-coercion of invoke/stream output to declared types.
Fixed time travel: Replays no longer reuse stale RESUME values; subgraphs correctly restore parent checkpoint.
Fully backwards compatible: version="v2" is opt-in.

Async subagents: Non-blocking background tasks while maintaining conversation.
Multi-modal support: read_file tool supports PDFs, audio, and video.
Backend changes: Updated file format for binary files, improved error propagation.

New sandbox integration packages: langchain-modal, langchain-daytona, langchain-runloop
Conversation history summarization now triggers on ContextOverflowError
Default to Responses API for OpenAI model strings

create_agent simplified support for provider-specific tool parameters via extras attribute
Support for strict schema-adherence in agent response_format via ProviderStrategy

Model profiles: Chat models expose supported features via .profile attribute
Summarization middleware: Flexible trigger points using model profiles
Model retry middleware: Configurable exponential backoff
Content moderation middleware: OpenAI-based unsafe content detection