Dia 1.6B
Nari Labs
Key facts · verified Oct 13, 2026
Dia 1.6B is a Nari Labs audio generation model released on 21 Apr 2025. Pricing: Open weights — free to self-host. Source: github.com.
Nari Labs' Dia, a 1.6-billion-parameter open-weight text-to-speech model under Apache 2.0, first released in April 2025. It generates two-speaker dialogue directly from a transcript tagged [S1] and [S2], including nonverbal sounds such as laughter and coughing, and can clone a voice from an audio prompt. It supports English. Nari Labs has since released Dia2 on GitHub and Hugging Face.
Pricing & specs checkedSource: vendor documentation · checked Oct 13, 2026LMArena: 1160 Elo (last listed Aug 22, 2026) · source · retrieved Aug 22, 2026
Key Features
Audio Generation Text-to-Speech
Full Specifications
Open weights — free to self-host | |
|---|---|
| Identity | |
| Developer | Nari Labs |
| Released | Apr 2025 |
| Licence | Apache 2.0 |
| Self-hostable | Yes — weights at nari-labs/Dia-1.6B |
| Capability | |
| Voice cloning | Yes — from an audio prompt and its transcript |
| Languages | English |
| Rights | |
| Commercial licence | Yes — Apache 2.0 |
| Measured quality | |
| LMArena Elo | 1160 (checked Aug 2026) |
| LMArena rank | Rank 12 |