Phonon Models by Gradium
Last refreshed 2026-10-20. Next refresh: weekly.
About
Phonon is Gradium's family of small on-device text-to-speech models, released October 20, 2026 in 31M and 40M sizes. The models stream 24 kHz speech on CPUs, GPUs and NPUs, build on Kyutai's Pocket TTS and Mimi codec, and are distilled from a larger Phonon teacher. Phonon 31M (English) is Apache 2.0; the 40M models (five single-language variants and a multilingual model) are CC BY-NC 4.0.
Decision facts
- Best fit
- text to speechcoding
- Capability starting point
- Phonon 31M
- Lowest tracked input
- Not tracked
- Closest related family
- 360Zhinao
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs text to speech, 31M parameters, and audio.
Use when the workload needs text to speech, 40M parameters, and audio.
Use when the workload needs text to speech, 40M parameters, and audio.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Phonon 31M | Use when the workload needs text to speech, 31M parameters, and audio. | 2026-10 | text to speech31M parametersaudio | Current |
| Phonon 40M | Use when the workload needs text to speech, 40M parameters, and audio. | 2026-10 | text to speech40M parametersaudio | Current |
| Phonon 40M Multilingual | Use when the workload needs text to speech, 40M parameters, and audio. | 2026-10 | text to speech40M parametersaudio | Current |
Release Timeline
1 release groupSpecifications(3 models)
| Model | Released | Parameters |
|---|---|---|
| Phonon 31M | 2026-10 | 31M |
| Phonon 40M | 2026-10 | 40M |
| Phonon 40M Multilingual | 2026-10 | 40M |