KyutaiVoice of Reason
GLM 4 Voice of Reason 9B
This research model adapts GLM 4 Voice 9B through reinforcement learning for spoken reasoning. An LLM judge supplies a binary reward for mathematical answers.
Research release
The training method changes the reasoning channel
The reported route uses reinforcement learning without an additional supervised fine-tuning stage. Speech generation still requires the upstream tokenizer and audio decoder; a language-model checkpoint alone is not the complete speech system.glm-4-voice-of-reason-9b: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/glm-4-voice-of-reason-9b |
| Upstream model | GLM 4 Voice 9B.glm-4-voice-of-reason-9b: Kyutai model card |
| Terms | The GLM 4 Voice license applies; this is not an Apache-licensed release.glm-4-voice-of-reason-9b: Kyutai model card |
Related models
- GLM 4 Voice of Reason STITCH 9B
Kyutai ยท Voice of Reason
A GLM 4 Voice 9B adaptation that combines silent text reasoning with spoken output and reinforcement learning.
Launch date not confirmedResearch releaseDownloadable