Audio
GA
Open weights
Moshi
Full-duplex spoken conversation research
Release2024-09
Context—
Max output—
ParametersUndisclosed
Local / VRAM16–24 GB
LicenseOpen weights
Overview
Moshi is a model in the Audio category from Kyutai. It was released in 2024-09. The vendor is based in France. Best for: Full-duplex spoken conversation research.
Notes: Positioning from the vendor site; specs, access and pricing can change — verify before production use.
Strengths
- Real-time speech in/out
- Open weights
- Low latency
Limits
- Research-grade
- Limited language support
Specifications
| Vendor | Kyutai |
|---|---|
| Category | Audio |
| Release | 2024-09 |
| Status | GA |
| License | Open weights |
| License name | CC-BY-4.0 |
| Commercial use | Not specified |
| Input modalities | Audio |
| Output modalities | Audio |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| VRAM | 16–24 GB |
| System RAM | 32–64 GB |
| API access | Yes |
| Pricing | Free (self-host) |
| Confidence | Medium |
| Vendor country | France |
| Vendor type | Research |
Pricing & hardware
Free (self-host)
This model can be self-hosted or run locally. Approximate GPU memory: 16–24 GB. Approximate system RAM: 32–64 GB. API access is available.