Haru-4B
Open full-duplex spoken-dialogue foundation model that listens and speaks simultaneously (7.5 Hz frames, 133 ms per frame) on a Qwen3.5-4B backbone; 9 languages; Apache 2.0.
Why it matters
Native full-duplex at phone-call latency with no separate audio transformer and no discrete audio tokens — a genuinely open base model for real-time voice conversation, #1 trending on Hugging Face in its first week.
What you could build with it
A startup could build a low-cost AI phone receptionist that interrupts, back-channels and responds in real time like a human operator, running Haru-4B on a single 12 GB GPU instead of paying per-minute hosted voice-agent fees.
Does it hold up?
Promising but explicitly a research base model, not an assistant: it holds natural conversation but is unreliable on facts and does not follow instructions, so builders should treat it as a voice layer to fine-tune, not a drop-in agent.
Built with Haru-4B
- Haru-4B Shows One Small Model Can Listen and Speak at Once, No ASR-LLM-TTS Stackmedium.com · Critical read of the 65 ms latency claim, the Moshi comparison, and what the release does and does not prove yet.
- Haru-4B Listens and Talks at the Same Time, Answering in About 65 Milliseconds on a MacBookwriteble.com · Coverage of the release, the 40M-hour pretraining corpus, and the model's stated limits.
Learn more
First spotted on hf: source.
More AI models
Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Qwen4-Max: Alibaba's new flagship model goes liveAlibaba's 2.4T-parameter flagship is now generally available via API, with a 1M-token context, native vision, and…models · JEV 0.67smolmaxxing: run 16B LLMs on a 1 GB TV box or a 2013 iPadAn MIT-licensed llama.cpp patch set that runs language models up to a 16B MoE on devices with about 1 GB of usable RAM…models · JEV 0.65Blue TTS v3: a 388MB open text-to-speech model with voice cloningAn open-source 388MB text-to-speech model (MIT) with voice cloning from about 10 seconds of speech, 7 languages, 15…models · JEV 0.65Ling 3.2 Flash: Ant Group's 560B native multimodal MoE, now free on OpenRouterAnt Group's inclusionAI launched Ling-3.2-flash, a 560B-parameter MoE (25B active per token) that reasons natively…models · JEV 0.63
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.