LangChain's anthropic 1.5.5 adds reasoning token tracking, fixes file-tool bug

LangChain, the popular AI framework with 144k GitHub stars, shipped version 1.5.5 of its Anthropic integration package, langchain-anthropic, on August 11. The release focuses on reliability and observability for developers using Claude models. The headline fix ensures reasoning tokens are now reported in usage metadata, giving teams accurate token counts for cost tracking and debugging—especially important for understanding how much of a model's context window is consumed by internal reasoning vs. visible output.

Beyond token reporting, the patch resolves a KeyError in Claude's file-tool middleware, a bug that could crash agent workflows when file-based tools were renamed. It also refreshes model profile data to keep performance and capability metadata in sync with the latest Anthropic models, and improves error messaging when no credentials are configured, making misconfigurations easier to diagnose. These are incremental but meaningful stability improvements for production systems relying on LangChain's Anthropic bindings.