LangChain 1.4.3 Patches Tool Calls, Adds Bedrock Mantle Support, and Fixes Cache Routing

What changed#

LangChain released version 1.4.3, delivering targeted patches across model initialization, agent tool execution, and dependency management. The framework now supports AWS Bedrock Mantle chat models through the init_chat_model utility, expanding cloud inference options for existing agent pipelines. A critical fix addresses how the library handles cache configurations when routing to fallback models, ensuring sanitized settings prevent downstream connection errors. Structured output parsing for GPT-6 has been corrected to work reliably without requiring explicit profile definitions. The most significant agent-specific update repairs invalid tool calls generated by create_agent, which previously broke multi-step reasoning loops. Documentation in AGENTS.md received updates to correct root setup guidance and improve package documentation accuracy. Internally, the team removed a stale commented-out Cohere extra flag and bumped the anyio network library from 4.11.0 to 4.14.2 within the /libs/langchain_v1 directory.

Why it matters#

These changes directly impact production-grade agent orchestration. The create_agent repair resolves a blocker where malformed tool instructions caused execution failures during complex planning sequences. Sanitizing cache settings for fallback models eliminates silent misconfigurations that often surface only after primary endpoints rate-limit or disconnect. Supporting Bedrock Mantle allows teams to route workloads across additional managed inference providers without custom wrapper code. The GPT-6 structured output fix reduces boilerplate when enforcing strict JSON schemas, while the anyio upgrade strengthens async HTTP client resilience under high concurrency. Together, these updates reduce debugging overhead for developers building stateful, multi-model agent systems.

What to do next#

Upgrade your environment using pip install langchain==1.4.3 or your preferred package manager. If you previously encountered tool execution drops inside create_agent workflows, verify that function signatures and parameter types align with the patched parser. Teams leveraging multiple inference providers should test the new init_chat_model routing logic against Bedrock Mantle endpoints. Review the revised AGENTS.md documentation if your deployment relies on the documented infrastructure setup patterns. Finally, confirm that your project resolves anyio >= 4.14.2 correctly, as the internal bump may affect transitive dependency resolution in stricter environments. The ReAct versus plan-and-execute comparison provides the architectural context for choosing an agent loop, while the n8n Agents update shows the same concerns in workflow automation.

This is a point release, but the affected paths are integration-sensitive. Validate the specific provider, cache fallback, and tool-call flows used by your application in CI; the release notes do not guarantee that an unrelated custom wrapper or deployment topology will behave identically.

Sources#

• Official LangChain 1.4.3 release notes