# Haru-4B Brings Full-Duplex Speech to Six Languages

> A compact 4-billion-parameter model handles real-time, two-way voice conversation across six languages in a single system.

Published by The Open Weights on Nov 12, 2026. Canonical: https://theopenweights.com/news/haru-4b-5ai5

## Key facts

- Company: Higurashi Ai
- Model: Haru-4B
- Version: 4B
- Category: Any-to-Any
- Modalities: Speech → Text, Any-to-Any, Text → Speech
- License: Other (Open weights, commercial use allowed)
- Parameters: 4B
- Architecture: HaruForConditionalGeneration
- Precision: BF16
- Weights size: 10.5 GB
- Languages: 6 languages
- Significance: notable
- Published: 2026-11-12
- Last verified: 2026-11-20
- Hugging Face: https://huggingface.co/higurashi-ai/Haru-4B
- Canonical URL: https://theopenweights.com/news/haru-4b-5ai5

Higurashi AI has released [Haru-4B](https://huggingface.co/higurashi-ai/Haru-4B), a compact speech-to-speech model built for full-duplex conversation. At roughly 4 billion parameters, it is designed to listen and speak at the same time rather than trading rigid turns, and it supports six languages out of the box.

Full-duplex is the detail worth dwelling on. Most voice assistants still operate in a strict turn-based loop: you talk, the system transcribes, generates, and then speaks back. A full-duplex approach lets the model handle overlapping speech, interruptions, and backchannel cues like "mm-hm" the way people actually converse — a meaningfully different interaction model from the familiar push-to-talk pattern.

## Why it matters

Speech pipelines have traditionally stitched together separate components for recognition, language modeling, and synthesis. Haru-4B folds those capabilities into one system, spanning:

- Speech-to-speech conversation as the primary mode
- Speech recognition (ASR) and synthesis (TTS)
- Coverage across six languages

At 4B parameters, the model sits in a size range that is realistic to run outside a data center, which lowers the barrier for developers building interactive voice interfaces. It is worth noting the model ships under a non-standard license, so teams should check the terms on the [model page](https://huggingface.co/higurashi-ai/Haru-4B) before building on it.

Details such as context length, supported sample rates, and the specific language list were not fully specified at release. For now, Haru-4B reads as an early but ambitious entry in the push toward natural, low-latency voice AI that runs at modest scale.

## Get the model

- [Hugging Face](https://huggingface.co/higurashi-ai/Haru-4B)

## Sources

- [higurashi-ai/Haru-4B](https://huggingface.co/higurashi-ai/Haru-4B) — Hugging Face, Nov 12, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Haru-4B Brings Full-Duplex Speech to Six Languages", The Open Weights, Nov 12, 2026, https://theopenweights.com/news/haru-4b-5ai5