Methods and apparatus for predictive head-tracked binaural audio rendering in which a rendering device renders multiple audio streams for different possible head locations based on head tracking data received from a headset, for example audio streams for the last known location and one or more predicted or possible locations, and transmits the multiple audio streams to the headset. The headset then selects and plays one of the audio streams that is closest to the actual head location based on current head tracking data. If none of the audio streams closely match the actual head location, two closest audio streams may be mixed. Transmitting multiple audio streams to the headset and selecting or mixing an audio stream on the headset may mitigate or eliminate perceived head tracking latency.