ICML 2026 | FusionRoute: From Expert Routing to Self-Correction in Multi-LLM Collaboration

ICML 2026｜FusionRoute：从专家路由到自我修正，一种新的多LLM协作范式

FusionRoute proposes a token-level multi-LLM collaboration method that freezes expert models and trains a lightweight router to select an expert for each token while merging router logits with expert logits. The paper evaluates it on GSM8K, MATH-500, HumanEval, MBPP, IfEval, and 500 PerfectBlend prompts.

Why it matters: HKR-H/K/R pass: token-level LLM routing is a strong research hook with concrete mechanics. The article lacks lift numbers, code link, and deployment cost, so it stays at the lower featured band.