Long-Horizon Reasoning

Long-Horizon Reasoning / Extended Planning

Definition

An LLM's ability to maintain coherent reasoning, planning, and decision-making over extended sequences of actions and time steps. Critical for agents that must operate autonomously for hours or days of simulated time.

Examples in the Wild

  • Example 1:An agent planning a multi-step strategy over 48 hours of world simulation
  • Example 2:Maintaining consistency in agent goals and values across hundreds of interactions
  • Example 3:Adapting strategy based on accumulated world state and prior outcomes
  • Example 4:Emergence World's evaluation of models across extended world-building scenarios