LangChain 1.5.0: agent platform drops Python 3.10 and adds a model router
On November 11, 2026, LangChain 1.5.0 was released — a framework for building agents and LLM applications that the authors call "the agent engineering platform." Six releases have accumulated since the last review: from 1.4.6 to 1.5.0.
What's new
- Dropping Python 3.10. The minimum supported version is now Python 3.11, and the minimum langchain-core version is 1.8.0. Users on Python 3.10 will continue to receive the 1.4.x branch.
- Model router and auto mode middleware. Middleware has been added for routing calls between models and automatically selecting a mode. ModelRouterMiddleware is marked as beta.
- Standardization of refusal content blocks. In langchain-core 1.7.5+, content blocks for model refusals have been unified — this simplifies handling situations where the model refuses to answer.
- Reading tool-call-delta in stream consumers. Stream consumers can now read tool call deltas — useful for agents that call tools in streaming mode.
- Per-call interrupts in HITL middleware. In 1.4.6, optional per-call interrupts were added to human-in-the-loop middleware.
- Middleware fixes. Internal model calls no longer end up in the messages stream, and trace inputs are excluded from ModelRefusalMiddleware.
What the platform does
LangChain is a framework for assembling agents and LLM applications from interchangeable components and third-party integrations. It provides a standard interface for models, embeddings, vector stores and tools. The platform supports RAG through connections to a variety of data sources, as well as tracing, quality evaluation and debugging through integrations such as LangSmith.
Your own model instead of the cloud
You can connect your own model through Ollama — the repository has a separate LangChain integration package for Ollama. The platform is installed with the command uv add langchain. License — MIT.
How it differs from its neighbors
Among comparable frameworks, LangChain is the only one whose documentation claims agent quality evaluation. At the same time, support for MCP, the A2A protocol, agent memory, human approval and stateful graphs is not mentioned in LangChain's documentation, whereas most of its neighbors claim these capabilities. All four platforms have their own tools and RAG.
| LangChain | CrewAI | Agno | AgentScope | |
|---|---|---|---|---|
| Type | framework | framework | framework | framework |
| Language | Python | Python | Python | Python |
| License | MIT | MIT | Apache-2.0 | Apache-2.0 |
| GitHub stars | 149,524 | 59,894 | 42,829 | 33,564 |
| Releases in 90 days | 15 | 23 | 27 | 6 |
| Multiple agents in a team | — | yes | — | yes |
| Own tools and functions | yes | yes | yes | yes |
| MCP support | — | yes | yes | yes |
| A2A protocol | — | yes | yes | yes |
| Agent memory | — | yes | yes | yes |
| Knowledge base (RAG) | yes | yes | yes | yes |
| Sandboxed code execution | — | — | — | yes |
| Human approval | — | yes | yes | — |
| Graphs and state persistence | — | yes | — | yes |
| Tracing and debugging | yes | — | yes | — |
| Agent quality evaluation | yes | — | — | — |
| Voice and speech | — | — | — | yes |
| Own server and Docker | — | — | yes | — |
| Images and documents | — | — | — | yes |
| Own model: Ollama | yes | yes | yes | yes |
| Own model: LM Studio | — | yes | — | — |
| Own model: OpenAI-compatible API | — | — | yes | — |
Checked against the projects' READMEs and documentation as of 11/15/2026: "yes" means the capability is stated there, a dash means it is not mentioned (which does not mean the capability is absent). Stars and releases are per GitHub data.
Those working on Python 3.10 should upgrade — it's time to move to 3.11 so as not to be left on the 1.4.x branch. Developers who care about agent quality evaluation and a mature integration ecosystem have good reason to take a closer look at the platform. You can follow news about running neural networks on your own hardware in the CPU3D news feed.
