Lab · moonshot
Orpheus-3B: specs & requirements
Orpheus is a Llama-3.2-based 3.8-billion-parameter speech model with emotion tags (<laugh>, <sigh>) and zero-shot cloning — top-tier quality served from GPUs. ONNX exports and a prototype JavaScript library exist, but its ~2.2 GB quantized weights out-of-memory most browsers. This page tracks it honestly: if it ever becomes runnable on enthusiast hardware, it lands in the Lab first.
Lab tier — heavy/frontier model for enthusiast hardware
Orpheus-3B support is being wired into the tool. Its specs and hardware verdicts are below — meanwhile, the live models are ready to use, and the compatibility check already tells you whether your device could run Orpheus-3B.
Specifications
| Developer | Canopy Labs |
|---|---|
| Parameters | 3800 million |
| Download size | 2230 MB (q4f16) |
| Output | 24.0 kHz mono |
| Voices | 8 |
| Languages | English |
| Voice cloning | Yes |
| Streaming | No |
| License | Apache-2.0 — Tagged Apache-2.0, but it is a Llama-3.2-3B derivative — Meta's Llama community license arguably still applies to derivatives. Do your own diligence for commercial use. |
| Speed | Unknown in-browser — nobody runs it there yet. Server-side it streams with ~200 ms latency on datacenter GPUs. |
| Runtime | WebGPU required |
Strengths
- +Highest raw quality of any open speech model with a theoretical browser path
- +Emotion tags for expressive delivery
- +Zero-shot voice cloning
Limitations
- −No working browser demo exists; ~2.2 GB weights need roughly 16 GB RAM + WebGPU
- −The promised smaller variants (150M/400M/1B) were never released
- −Llama-derivative license ambiguity
Orpheus-3B — questions
Is Orpheus-3B free to use?
Yes. The weights are Apache-2.0 and they run on your device. Tagged Apache-2.0, but it is a Llama-3.2-3B derivative — Meta's Llama community license arguably still applies to derivatives. Do your own diligence for commercial use.
What hardware does Orpheus-3B need?
A WebGPU-capable browser and GPU (Chrome/Edge, Safari 26+, Firefox 141+) plus about 16 GB of RAM. It will not run on CPU at usable speed.
Is my text private when using Orpheus-3B here?
Yes. The model executes inside your browser tab — the text you synthesize is never transmitted to a server.
Can I use Orpheus-3B output commercially?
Review the current model, voice, and source-dataset terms before commercial use. The catalog lists Apache-2.0 for this model and notes: Tagged Apache-2.0, but it is a Llama-3.2-3B derivative — Meta's Llama community license arguably still applies to derivatives. Do your own diligence for commercial use.