Certain aspects of the present disclosure provide techniques for packet buffering. A method that may be performed by a receiving node includes dynamically determining one or more time durations to buffer packets. The one or more time durations can be different than a time duration of a configured timer for buffering the packets. The receiving node may input one or more parameters to a machine learning algorithm and obtain, as output of the machine learning algorithm based on the input one or more parameters, one or more time durations to buffer packets. The receiving node buffers packets for the determined one or more time durations. The receiving node may use machine learning to dynamically determine the one or more time durations to buffer packet. The buffering may be at a radio link control (RLC) reassembling buffer and/or a packet data convergence protocol (PDCP) buffer.