OpenAI
Go SDK
The SDK continued rapid expansion of multimodal and agentic capabilities. Custom voices shipped in v3.28.0 with a breaking change to the voice parameter structure, while video editing and extension APIs rolled out alongside Sora character support in v3.27.0. Concurrently, the SDK surfaced new tool types—computer use graduated from preview, tool search and skills appeared, and batch processing expanded to handle images. Short-lived token support landed in v3.31.0, and the client gained a StreamError type to access raw SSE events directly rather than only parsed results.
The SDK expanded computer use and video capabilities while aligning type definitions across the API. Custom voices landed with a breaking change to the voice parameter—now accepting either a string or an object with an id field—alongside new video editing and character generation endpoints. Model support added gpt-5.4 and nano/mini variants, function tools gained defer_loading, and batch requests now support the /v1/videos endpoint.