higurashi-ai has released Haru-4B, a multilingual, full-duplex conversational AI model designed for real-time spoken dialogue. The model can listen and speak simultaneously, processing audio and generating text at a rate of 7.5 Hz, with each step taking 133 milliseconds. Haru-4B supports character and voice control, allowing users to customize its persona and speech patterns using reference clips, and is available under the Apache License 2.0 for research purposes. AI
IMPACT Enables more natural, real-time spoken interactions by allowing simultaneous listening and speaking.
RANK_REASON Model release from a research lab with technical details. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →