single mention · model architecture
Thinker–Talker architecture
Two stacks: the Thinker is a language model that reasons and writes text, and the Talker is a second model that turns that text and context into speech codec tokens. In Qwen3.5-Omni the Talker works on one interleaved stream of text and speech tokens.
Written for the explainer Hearing, speaking and keeping time: audio, omni models and multimodal positions.
Filed under model architecture › multimodal architecture