Japanese startup open-sources Haru-4B, a full-duplex voice AI that tops HF trending
solyarisoftware · x · 2026-11-20
Japanese startup higurashi AI open-sourced Haru-4B, a voice model built for full-duplex conversation: it listens while speaking, handles interruptions, generates overlapping speech, gives backchannel acknowledgments, and converses across 9 languages. It hit #1 on Hugging Face's trending models list across all categories.
- Architecture based on Qwen3.5-4B plus audio components; runs locally on an NVIDIA GPU with roughly 12GB VRAM
- Developers report 65.2ms median processing latency and one audio frame every 133ms on an Apple M1 Max, using custom Metal inference kernels
Related event: Japanese Startup Open-Sources Full-Duplex Speech Model Haru-4B(3 posts)→
More from Models
- Timed eval task asks users to deliberately make Claude fail at real work in 12 minutes — Icy-Measurement10 · 2026-11-22
- Tesla FSD 15's real-time view isn't 3D rendering — it's the AI model's own synthetic perception — mark_k · 2026-11-22
- Gemini Black Friday half-price: JP-region trick gets Pro annual plan for $91.54 — huangyun_122 · 2026-11-22
- Microsoft's USB4 RDMA lets mini PCs pool memory to run Qwen3-235B — JFPuget · 2026-11-22
- Google AI Mode now enforces usage limits, users surprised by rate-limit popup — elvano124 · 2026-11-22
- Sol 6.2 now refuses reverse engineering that Sol 6.1 happily did for a month — cardak98 · 2026-11-22