PersonaPlex 7B

PersonaPlex-7B-v1 is NVIDIA's most talked-about model release of early 2026. Released on January 15, 2026, it is a 7-billion parameter full-duplex speech-to-speech conversational AI — the first open-source voice model at this scale capable of listening and speaking simultaneously. It is part of the broader NVIDIA Open Model Ecosystem and available free on Hugging Face and GitHub. [NVIDIA AI Models 2026 Guide]

Most voice assistants operate like walkie-talkies: the user speaks, the system waits, processes, then responds. PersonaPlex eliminates this pattern entirely. [NVIDIA AI Models 2026 Guide]

PersonaPlex is built on the Moshi architecture, using a dual-stream Transformer that processes incoming user audio and generates response audio at the same time. There is no handoff between separate ASR, LLM, and TTS systems — the entire pipeline collapses into a single model. [NVIDIA AI Models 2026 Guide]

On FullDuplexBench, PersonaPlex outperforms Gemini Live, Qwen 2.5 Omni, and Moshi on both conversational dynamics and task adherence. [[src:NVIDIA AI Models 2026 Guide]]

## Persona Control

Before a conversation begins, PersonaPlex is conditioned on two inputs: [[src:NVIDIA AI Models 2026 Guide]]

| Input | What it controls |
|---|---|
| **Voice prompt** (audio tokens) | Tone, accent, and speaking style |
| **Text prompt** | Role, background, and scenario context |

This allows the model to portray a wide variety of characters — from a wise teacher to a bank customer service agent — while maintaining that persona throughout the entire conversation, **including when interrupted**. [[src:NVIDIA AI Models 2026 Guide]]

## Architecture

PersonaPlex is based on the **Moshi architecture** with a dual-stream Transformer design. The architecture unifies what would traditionally be three separate systems (ASR, LLM, TTS) into a single model that updates its internal state continuously as the user speaks and streams a response back immediately. [[src:NVIDIA AI Models 2026 Guide]]

## Licensing & Availability

- **Model weights**: NVIDIA Open Model License (commercial use permitted, no royalties)
- **Code**: MIT License (commercial use permitted)
- **Hugging Face**: `nvidia/personaplex-7b-v1`
- **GitHub**: `NVIDIA/personaplex`
- **Also listed on**: [[nvidia-build|NVIDIA Build]] alongside other free models [[src:NVIDIA AI Models 2026 Guide]]

For API-based access without managing GPU infrastructure, NVIDIA offers deployment through [[nvidia-nim|NVIDIA NIM (Inference Microservices)]]. [[src:NVIDIA AI Models 2026 Guide]]

## Hardware Requirements

PersonaPlex-7B-v1 requires **NVIDIA Ampere or Hopper architecture GPUs** for real-time performance. [[src:NVIDIA AI Models 2026 Guide]]

NVIDIA has confirmed sub-second latency only on data center-class hardware. [NVIDIA AI Models 2026 Guide]

PersonaPlex is framed not just as a model release but as a market structure event. When an open-source 7B model running on a single GPU can match or exceed commercial voice APIs, it significantly erodes the pricing power of commercial voice AI providers. As noted in the source: NVIDIA effectively commoditized the voice AI stack overnight. [NVIDIA AI Models 2026 Guide]

This follows the same hardware-adoption logic seen across the NVIDIA Open Model Ecosystem: free software, paid hardware. The model itself costs nothing; the requirement for high-end NVIDIA GPUs ensures that widespread adoption drives hardware revenue. This strategy is also discussed in Free vs. Paid AI Model Access: NVIDIA NIM vs. $20/month Subscriptions. [NVIDIA AI Models 2026 Guide]

PersonaPlex makes it trivially easy to clone voices and create human-sounding AI agents with arbitrary personas. The source flags this as an enormous potential for scams, social engineering, and disinformation. While NVIDIA includes safety guidelines, the open-source nature of the model means guardrails can be removed by anyone. This is identified as a genuine societal risk that the industry may be underestimating. [NVIDIA AI Models 2026 Guide]

PersonaPlex outperforms all three on FullDuplexBench for conversational dynamics and task adherence. [NVIDIA AI Models 2026 Guide]

• NVIDIA Open Model Ecosystem — the broader portfolio PersonaPlex belongs to
• NVIDIA NIM (Inference Microservices) — managed deployment option for PersonaPlex
• NVIDIA Build — platform where PersonaPlex and other free models are listed
• Rubin Hardware Platform — next-gen hardware that will reduce inference costs for models like PersonaPlex
• Nemotron 3 — NVIDIA's flagship LLM family, a sibling model line
• NVIDIA vs. OpenAI, Google, AMD & Meta: AI Competitive Landscape 2026 — broader competitive context