Reference · server-class
MOSS-TTS (8.5B): specs & requirements
MOSS-TTS (and its v1.5 revision) is OpenMOSS's 8.5-billion-parameter flagship — state-of-the-art cloning and expressiveness. We document it here for honesty: at ~4.8 GB quantized it exceeds what any browser runtime can execute (WebAssembly is capped at 4 GB, and real inference would need 7–9 GB of tab memory). There is no browser export, and unlike our Lab models, none is plausible. If you want MOSS technology in your browser, MOSS-TTS-Nano is the official 100M browser member of the same family; if you want the flagship, run it server-side from the linked repo. The mid-size family members (Realtime 2.3B, Local-Transformer 3–4.6B) also have no browser exports today — we'll move any of them into the Lab if that changes.
Reference — not browser-runnable
This model physically exceeds what browsers can execute, so unlike our Lab models it has no planned adapter — this page exists so you get a straight answer instead of a fake demo. If you want MOSS technology here, use MOSS-TTS-Nano, the official in-browser member of the same family — no setup, no server.
Specifications
| Developer | OpenMOSS / MOSI.AI (Fudan) |
|---|---|
| Parameters | 8490 million |
| Download size | 4800 MB (q4 GGUF (server-side)) |
| Output | 48.0 kHz mono |
| Voices | 1 |
| Languages | Chinese, Cantonese, English, Arabic, Czech, Danish, German, Dutch, and more (v1.5 multilingual) |
| Voice cloning | Yes |
| Streaming | Yes — audio starts before generation finishes |
| License | Apache-2.0 |
| Speed | Server GPUs only — no browser runtime can execute an 8.5B speech model today. |
| Runtime | WebGPU required |
Source: OpenMOSS-Team/MOSS-TTS-v1.5
Strengths
- +State-of-the-art open TTS quality (server-side)
- +Same Apache-2.0 family and tokenizer lineage as MOSS-TTS-Nano
- +48 kHz output, cloning, and streaming — on datacenter hardware
Limitations
- −Cannot run in a browser — ~4.8 GB quantized weights vs a 4 GB WebAssembly ceiling
- −No ONNX/browser export exists for any MOSS model above Nano
MOSS-TTS (8.5B) — questions
Is MOSS-TTS (8.5B) free to use?
Yes. The weights are Apache-2.0 and they run on your device.
What hardware does MOSS-TTS (8.5B) need?
Server-class hardware — a datacenter or workstation GPU. No browser can run it: its quantized weights alone exceed the 4 GB WebAssembly ceiling, and no browser export exists.
Is my text private when using MOSS-TTS (8.5B) here?
Yes. The model executes inside your browser tab — the text you synthesize is never transmitted to a server.
Can I use MOSS-TTS (8.5B) output commercially?
Review the current model, voice, and source-dataset terms before commercial use. The catalog lists Apache-2.0 for this model.