Tag
Introduces SR²AM, a framework for efficient agentic reasoning via self-regulated simulative planning, achieving competitive performance with models 20-30x larger while using 26-95% fewer reasoning tokens.