← Back to Voice models
NVIDIA
NemotronLabs VoiceChat 11B
Released Aug 3, 2026
NVIDIA's open-weight full-duplex speech model, an 11B hybrid Mamba/Transformer. NVIDIA calls it the first open full-duplex model that keeps the conversation going while it calls tools.
- Price per minute
- —
- Pricing
- Open weights; no hosted price.
- Inputs
- audio, text
- Tool calling
- Yes
- Languages
- 1
- Voices
- —
- Big Bench Audio
- 27%
- Access
- Generally available
- Licence
- OpenMDW 1.1
Strengths & weaknesses
Strengths
- Open weights under OpenMDW 1.1
- Tool calling with spoken 'on-hold' messages while tools run
- 448 ms turn-taking latency on Full-Duplex-Bench 1.0 (NVIDIA)
Weaknesses
- English only
- Big Bench Audio 27.0% (AA)
- No hosted API: self-serve via vLLM or an NGC NIM container
Not notable: #41 on AA Big Bench Audio (27.0%, listed as 'Nemotron Voicechat'). Release date conflict: the model card says 2026-08-03, the NGC container page 2026-07-21. Licence conflict: the HF card says OpenMDW 1.1, but the NGC NIM container is under the NVIDIA Software and Model Evaluation License; the HF weights licence is used here. NVIDIA reports BFCL-v3 spoken tool calling 56.1%.
Links
API model strings
- huggingface
nvidia/NVIDIA-NemotronLabs-VoiceChat-11B