The langchain-openai package has matured rapidly from its initial release through version 1.5.2a1, delivering a wide range of new capabilities for developers building AI applications on top of OpenAI’s services. Major feature additions include support for the OpenAI Responses API, which enables more flexible interaction patterns, as well as compatibility with the latest OpenAI SDK versions (2.0 and 3.0). The library now supports o-series reasoning models, including token counting and streaming, and introduces new features such as explicit prompt caching, the apply_patch built-in tool, tool search, server-side compaction, and configurable reasoning effort and verbosity. Developers can also leverage a new ChatGPT OAuth-backed chat model (ChatOpenAICodex), support for structured output and JSON schema, and automatic gateway metadata extraction from response headers.
Beyond new features, the package has seen extensive bug fixes and compatibility improvements across its lifespan. Notable fixes include correct handling of token counting for various content types, streaming hang prevention, filtering of invalid tool calls, and preservation of streamed encrypted reasoning. Azure OpenAI integration received numerous improvements, such as better validation for API base URLs, streaming token counts, and support for the Responses API. The codebase has been updated to support Python 3.13 and 3.14, Pydantic v2, and a wide array of dependency updates (pillow, openai, langgraph, langsmith, urllib3, and others). Security enhancements include SSRF-safe transport for image token counting, redaction of MCP authorization headers, and addressing a Pygments vulnerability. Performance was also improved through async tokenization, caching of httpx clients, and optimized embedding calculations.
Documentation and developer experience also received substantial attention. Docstrings were standardized and clarified across the package, the README was refreshed, and API reference documentation was enhanced. The project migrated its tooling to uv and hatchling, improved CI workflows, and added extensive test coverage, including standard tests, VCR cassettes, and integration tests for chat, embeddings, and structured output. These changes make the langchain-openai package more reliable, easier to maintain, and better documented for self-hosted users and enterprise deployments.
