langchain-core==1.2.15

LangChain AI has pushed a new maintenance release for its foundational library, langchain-core version 1.2.15. This patch focuses on developer experience, performance, and documentation clarity rather than introducing new features.

The key technical improvement is a performance optimization (PR #35298) that defers the import of specific LangSmith SDK modules. LangSmith is LangChain's observability and evaluation platform. By making these imports lazy, the overall import time for the core library is reduced, leading to faster startup for applications that don't immediately use LangSmith tracing. Another significant fix improves the error message users receive when passing a non-JSON-serializable object as a tool schema (PR #34376), making debugging custom tools and agents more straightforward. The release also includes a documentation and typing update for the `on_chat_model_start` callback to explicitly clarify its required positional arguments (PR #35324), preventing runtime errors.

Contextually, this release follows the recent addition and subsequent reversion of a `ChatAnthropicBedrockWrapper` in the previous version (1.2.14). Version 1.2.15 properly re-introduces this wrapper (PR #35091) for using Anthropic's Claude models on AWS Bedrock. It also updates test code and docstrings to replace retired Anthropic model IDs (PR #35365), ensuring examples remain functional.

The practical implication is a smoother, slightly faster development cycle for engineers building LLM applications with agents, RAG systems, and complex chains. The changes are incremental but address specific pain points around import bloat and opaque error messages, reinforcing the stability of the widely adopted framework.