The latest update to the LangChain-Anthropic integration, version 1.5.0, brings a notable new feature and important fixes for self-hosted users. This release focuses on enhancing model controllability and ensuring smoother tool interactions.
Reasoning Effort Control: A new standard parameter, reasoning_effort, has been added to chat models. This allows developers to adjust the depth of reasoning the model applies before generating responses. Whether you need quick, low‑latency answers or deeply considered outputs, you can now tune this behavior directly through the chat model interface.
Tool Recognition Fix: A bug where built‑in tools with an “advisor_” prefix were not properly recognized has been patched. This ensures that all built-in tools function correctly, preventing unexpected failures when using advisory tools in your LangChain applications.
Alongside these functional changes, the release includes routine maintenance such as refreshed model‑profile data, updated dependency lockfiles, and a docstring update to reflect the recent Sonnet‑5 release. Additionally, testing hygiene was improved by ignoring LangSmith requests in VCR cassettes, which makes local testing more reliable. These updates keep the integration current and robust for all users.
