OpenAI
Node SDK
The SDK continues expanding its real-time and multimodal capabilities while stabilizing core APIs. The team shipped support for WebSocket streaming in the Responses API, added short-lived token authentication, and introduced async iterators for WebSocket classes to improve streaming ergonomics. Simultaneously, deeper integrations landed across Sora (video character API and higher-resolution exports), computer use (graduating from preview with keyboard input support), custom voices, and batch processing of images. Type refinements addressed schema misalignments in response handling, and newer model variants like gpt-5.4 and gpt-realtime-1.5 were added to the client.
WebSocket streaming and computer vision capabilities dominated March. The SDK added async iterator and stream() methods to WebSocket classes for cleaner consumption patterns, while shipping gpt-5.4, a general availability ComputerTool (graduated from preview), and new tool search functionality. Sora video API improvements landed with character controls, video extensions and edits, and higher-resolution exports. Type correctness improved across response schemas, including making the type field required in ResponseInputMessageItem.