PANews reported on February 25th that, according to OpenAI developer documentation, OpenAI has introduced a WebSocket mode for its Responses API, specifically optimized for complex workflows involving frequent tool calls. This mode, by establishing persistent connections and supporting incremental input, can improve the execution speed of long-chain tasks containing more than 20 tool calls by approximately 40%. Furthermore, the WebSocket mode is compatible with the Zero Data Retention (ZDR) specification, supporting low-latency context reconnection using previous_response_id. Currently, the duration of a single connection is limited to 60 minutes.


