OpenAI
Python SDK
The SDK shipped 16 releases over 90 days, steadily expanding support for new APIs and runtime capabilities. The major themes were adding structured APIs for agent execution (computer use, web search, image generation, shell commands), enabling real-time communication patterns (websockets for responses, short-lived tokens), and supporting emerging model capabilities (GPT-5.4, custom voices, Sora video extensions). Notably, v2.26.0 graduated computer use from preview to general availability, and v2.31.0 added websocket support for raw binary data alongside new conversation messaging fields and query serialization improvements for complex parameter formats.
March focused on expanding the computer use and video capabilities alongside new model releases. The SDK added keys field support to computer actions and graduated ComputerTool from preview to general availability, while separately shipping GPT-5.4 model variants, Sora video extensions and character API support, and a new tool search capability. Response type alignment fixes and schema updates landed throughout, including making type required in ResponseInputMessageItem and adding in/nin operators to comparison filters.