← Back to Voice models
NVIDIA
PersonaPlex-7B
Released Jan 15, 2026
NVIDIA's open-weight 7B full-duplex speech-to-speech model, built on Kyutai's Moshi. A text prompt sets the persona and an audio prompt sets the voice.
- Price per minute
- —
- Pricing
- Open weights; no hosted price.
- Inputs
- audio, text
- Tool calling
- No
- Languages
- 1
- Voices
- —
- Big Bench Audio
- 19.1%
- Access
- Generally available
- Licence
- NVIDIA Open Model License Agreement
Strengths & weaknesses
Strengths
- Persona set by text prompt, voice by audio conditioning
- Listens and speaks at once, with barge-in
- Open weights, ready for commercial use per NVIDIA
Weaknesses
- English only
- Big Bench Audio 19.1% (AA)
- Gated Hugging Face download; self-host only
Not notable: #42 on AA Big Bench Audio (19.1%). Release date per the model card (2026-01-15); secondary coverage says 2026-01-20. Licence: NVIDIA Open Model License Agreement, with additional information under CC-BY-4.0. The GitHub repo ships pre-packaged voice prompts (NATF0-3, NATM0-3, VARF0-4, VARM0-4), but the model card gives no count, so voices is null.
Links
API model strings
- huggingface
nvidia/personaplex-7b-v1