The latest patch release for the LangChain-OpenAI integration brings a focused set of fixes that improve reliability and security for self-hosted users. A new ContextWindowExceededError is now properly raised when the model’s token limit is breached, allowing your application to gracefully handle oversized prompts rather than failing silently. Additionally, the upgrade preserves custom text options—like temperature and top_p—in Responses API calls, so your carefully tuned generation parameters aren’t unexpectedly discarded.
Security and data hygiene also see attention: authorization headers in MCP connections are now redacted from logs, preventing accidental exposure of sensitive credentials. Internally generated content block IDs are filtered out of model outputs, keeping the response stream clean and free of confusing artifacts. On the maintenance front, a refresh of model profile data ensures compatibility with the latest OpenAI offerings, while several dependency bumps across the codebase maintain overall stability.
