LangChain 1.5 adds decision models and router middleware, and drops Python 3.10 (November 11, 2026)

What changed

On November 11, 2026 LangChain released langchain 1.5.0. It needs langchain-core 1.8.0 or later. Both drop Python 3.10, so you need Python 3.11 or later. On Python 3.10, pip keeps resolving langchain 1.4.x.

The new langchain_core.decisions module adds BaseDecisionModel for typed, probabilistic answers: yes/no (Noul), choice (Choice), score (Score) and bounding box (Box).

Two middleware, released earlier in langchain-typesafe, now live in langchain.agents.middleware, marked beta:

• ModelRouterMiddleware picks a model for each run from your configured choices and instructions.
• AutoModeMiddleware classifies configured tool calls and blocks the ones it judges risky.

Both now require an explicit decision model passed as classifier=. The experimental imports in langchain-typesafe are deprecated; TypeSafe 0.1 ships TypeSafeClassifier for this.

SummarizationMiddleware now accepts a separate model or a custom callable through summarizer=.

Prerequisites

• Python 3.11 or later
• A Python app that uses langchain
• No API key is needed for the upgrade itself

Steps

1. Check your Python version with python --version. Move to 3.11 or later if needed, including CI, Docker images and deploy runtimes.
2. Upgrade: pip install -U "langchain>=1.5.0" "langchain-core>=1.8.0".
3. If you used the langchain-typesafe experimental middleware, import it from langchain.agents.middleware instead and pass classifier=TypeSafeClassifier().

Test it

Run pip show langchain and check it reports 1.5.0. Then start your agent with the middleware imported from langchain.agents.middleware and classifier= set: it should start without a deprecation warning.

Not confirmed

• Which providers besides TypeSafe ship a BaseDecisionModel.