AI · November 15, 2026 · Moderate impact
Haru-4B releases an open full-duplex model that listens and speaks at once
The higurashi-ai team has released Haru-4B, a pretrained base model for low-latency multilingual full-duplex spoken dialogue, available for research. The model works in 133 ms steps at 7.5 Hz: each step takes one frame of user audio and produces one frame of its own audio and text. User audio, model audio and model text are summed into a single position of the language model, so listening and speaking never wait for each other. Code and a technical blog, including a Japanese version, accompany the weights.
It gives researchers an open base for phone-call style voice interaction without turn-taking delays.