Lost in Context: Adressing Context Anxiety in Large Language Models

Conventional wisdom suggests that reasoning models fail when problems exceed their capabilities. However, we find that frontier reasoning models sometimes possess the necessary capabilities to solve problems but fail due to premature self-doubt -- a phenomenon informally known as context anxiety. We provide the first systematic study of context anxiety, demonstrating that it arises, in part, from a model's inability to accurately estimate the tokens required to complete a task. We also show that context anxiety leads to material efficiency losses when models operate under perceived constraints. Building on this analysis, we further show that models can learn alternative strategies for solving long-horizon problems without exhibiting context anxiety, suggesting that performance improvements may be achievable not through scaling model capabilities, but by improving models' ability to accurately assess and adapt to their own limitations.

When AI models fail at complex tasks, we typically assume the problem was just too hard for them. However, we discover that these models often possess the right skills but give up too early -- a phenomenon called context anxiety. This self-doubt occurs because the AI struggles to accurately estimate how much time and effort a task will actually require. To tackle this, we conduct the first systematic study of context anxiety, showing how it causes AI to waste resources and underperform when feeling constrained. We then show that models can be taught new strategies to confidently tackle long, complicated problems without experiencing this anxiety. Our research suggests a promising new direction for the future of AI. Instead of constantly making models larger and more computationally expensive, we can achieve major performance improvements simply by helping them better understand, assess, and adapt to their own limitations.