OpenAI has introduced WebSocket support for its Responses API, significantly enhancing the speed of complex workflows. This new mode, detailed in OpenAI's developer documentation, is optimized for tasks involving frequent tool calls, improving execution speed by approximately 40% for long-chain tasks with over 20 tool calls. The WebSocket mode establishes persistent connections and supports incremental input, while also being compatible with the Zero Data Retention (ZDR) specification for low-latency context reconnection. Each connection is currently limited to a duration of 60 minutes.