# LangGraph 1.2.18 takes delta snapshots every 100 updates by default

published: 2026-10-26T20:05:31.043Z
topic: frameworks
source: https://github.com/langchain-ai/langgraph/releases/tag/1.2.18
url: https://aiagentnews.io/stories/langgraph-1-2-18-takes-delta-snapshots-every-100-updates-by-default

LangGraph 1.2.18 is out, with fixes to error handling, streaming and subgraph tags. Its main behavior change is that DeltaChannel now takes a full snapshot every 100 updates instead of every 1000.

LangGraph, the agent framework from LangChain, published version 1.2.18 on GitHub. The release changes the default for DeltaChannel, which now writes a full snapshot every 100 updates rather than every 1000. According to the changelog, that makes reads faster on long threads at the cost of more storage, and developers who want the old behavior can pass snapshot_frequency=1000.

Other fixes let error_handler recover on every path and stream the results of failed tasks. Runs stopped by interrupt_after or the recursion limit now stream their last checkpoint. Tags are also no longer duplicated at each subgraph level.

Dependency updates bump opentelemetry-api from 1.39.1 to 1.45.1 and nbconvert from 7.17.1 to 7.17.2 in libs/langgraph.

## Why it matters if you build agents

If your agents keep long threads on DeltaChannel, expect full snapshots to arrive more often, so storage use will grow while reads get faster. Set snapshot_frequency=1000 if you want to keep the previous write pattern.

Code that consumes streams from runs halted by an interrupt or the recursion limit will now receive the final checkpoint, so test any logic that assumed it would not.