AI RESEARCH
Multi-objective Evolutionary Merging Enables Efficient Reasoning Models
arXiv CS.CL
•
ArXi:2604.06465v1 Announce Type: new Reasoning models have nstrated remarkable capabilities in solving complex problems by leveraging long chains of thought. However, this deliberate reasoning comes with substantial computational overhead at inference time. The Long-to-Short (L2S) reasoning problem seeks to maintain high accuracy using fewer tokens, but current