NemotronLabs VoiceChat 11B
Pandan is the first inference provider for NVIDIA NemotronLabs VoiceChat 11B, the first open full-duplex model. It listens and speaks at the same time, handles interruptions, and calls tools.
ReadyTurn on your microphone to start a conversation.
Turn on your microphone, wait for the model to report Live, then speak normally. You can interrupt it in the middle of a response.
Use headphones for the cleanest full-duplex test.
448 msTurn-taking latency
480 msInterruption latency
#1VoiceBench full-duplex
116 msPer 160 ms audio step
Latency figures are NVIDIA-reported. The VoiceBench leaderboard ranks Nemotron 3 VoiceChat (V1) #1 for full-duplex speech-to-speech. In a 115-second live H100 session, Pandan measured 116 ms of inference per 160 ms audio step.
Research preview · English · fixed voice · two-minute sessions · model released August 3, 2026