ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces

Open paper

Benchmark	Model	Rank	Results
code-generation-on-humaneval	ThinkRouter (Qwen3-1.7B)	–	Pass@1: 81.82
code-generation-on-mbpp	ThinkRouter (Qwen3-32B)	–	Accuracy: 96.49