LangChain v1.1.0 is now available

This release makes agent development more reliable, more structured, and more context-aware — powered by a new model-profile system and expanded middleware.

Model profiles (new)

Chat models now expose a .profile attribute describing supported capabilities (structured output, function calling, JSON modes, etc.).
These profiles are sourced from models.dev, the open-source project indexing model behaviors across providers — giving developers an accurate, declarative surface to build on.

Smarter, context-aware summarization middleware

The summarization middleware now uses model profiles to decide when and how to summarize, providing flexible trigger point, provider-specific behavior, and better performance on long-running agent sessions

Structured output improvements

ProviderStrategy (native structured output) can now be inferred directly from model profiles — making it easier to build agents that return reliable, typed results without hand-coding provider logic.

SystemMessage support in create_agent

create_agent now accepts SystemMessage instances via system_prompt, enabling:

• cache-control blocks

• structured orchestration hints

• richer system-level instructions

• compatibility with advanced content blocks

New model-retry middleware

We’ve added built-in retry logic with configurable exponential backoff to improve resilience across flaky model endpoints. This ensures agents gracefully recover from transient provider errors.

Content moderation middleware

OpenAI moderation can now be applied as middleware across user input, model responses, and tool-returned content — giving teams a consistent safety layer for agent interactions.

Learn more: https://docs.langchain.com/oss/python/releases/changelog