Skip to content
Agent Radar

langchain-core 1.7 streams tool-call args as deltas

Updated LangChain

Summary

langchain-core 1.7.0 streams tool-call args as tool-call-delta events; 1.7.1 (October 28, 2026) fixes gateway tracing metadata and Mermaid labels.

Tool
LangChain
Category
Frameworks
Date
Open source
Yes
Pricing
Not stated in the sources we cited.
Sources

langchain-core 1.7.1 was published on October 28, 2026, six days after 1.7.0. The minor bump is the part to read: streamed tool-call arguments now arrive as tool-call-delta slices, which changes what raw stream consumers see. 1.7.1 is a small fix release on top.

Release facts

FactDetail
Packagelangchain-core, the base package under LangChain
Versions1.7.0 (October 22, 2026) and 1.7.1 (October 28, 2026)
LicenseMIT, per the GitHub API
Installpip install -U langchain-core
Who is affectedCode that reads raw chat-model stream events while a tool call is being generated

The change in 1.7.0

  • Tool-call args stream as deltas. Before 1.7.0, streamed tool-call arguments came as block-delta events holding a snapshot of the args so far. From 1.7.0 they come as tool-call-delta events holding only the new slice (PR 41257).
  • What stays the same. The release notes say .tool_calls.get() and .output are unchanged, so code that waits for the finished message is not affected.
  • Groundwork in 1.6.13. The stream consumers in core, langchain and the standard tests were taught to read tool-call-delta in 1.6.13 (October 21), and since that version iterating ChatModelStream.tool_calls yields each call's new args slice rather than a snapshot.
  • Escape hatch. The notes say to pin langchain-core<1.7 if you need the old snapshot behaviour.

What 1.7.1 fixes

  • Gateway tracing metadata survives v3 streams (PR 41504), so traces keep the metadata a model gateway attached.
  • Mermaid graph labels are escaped (PR 41512): node metadata with special characters no longer breaks a rendered Mermaid diagram.
  • Looser dependency caps. Third-party upper bounds were relaxed, plus routine dependency bumps.

Questions

Do I need to change code for langchain-core 1.7? Only if you parse raw stream events and read tool-call args from block-delta mid-stream. Switch that code to tool-call-delta and accumulate the slices yourself. If you only use the final message, .tool_calls.get() or .output, nothing changes.

Should I take 1.7.1 if I am already on 1.7.0? Yes. It is a patch with two fixes and dependency updates, and the release notes list no new behaviour change.

How do I keep the old snapshot behaviour? Pin langchain-core<1.7. Note that iterating ChatModelStream.tool_calls already yields slices from 1.6.13, so pin below that if your code relies on snapshots there.

More on the LangChain hub, the earlier langchain-anthropic 1.7.6 launch and the frameworks category. If you are weighing a switch, see LangChain alternatives.

Alternatives

LangChain compared with its alternatives
ToolLicensePricing note
LangChainThis launchLangChain is an open-source agent engineering platform and Python framework for building agents.
MITNot stated
LangGraphLangGraph is an open-source orchestration runtime for building stateful, resilient agents.
MITNot stated
Pydantic AIPydantic AI is a model-agnostic Python framework for building typed AI agents and applications.
MITNot stated
CrewAICrewAI is a framework for orchestrating role-playing, autonomous AI agents that work together.
MITNot stated
MastraTypeScript agent framework. Mastra Code is the terminal agent; AgentController is the extracted loop, in @mastra/core 1.47.0 or later.
Open sourceMastra Code and AgentController are announced as open source. No license fee is stated in the posts we cited.

More from LangChain

View all

Questions

What is langchain?

The GitHub repository describes langchain as "The agent engineering platform."

Is langchain open source?

The GitHub API reports an MIT license for langchain-ai/langchain.

How do I create an agent with langchain?

The docs overview shows importing create_agent from langchain.agents and passing a model, tools and a system prompt.

Filed on Agent Radar. Claims in this note are limited to the sources above.