FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning

ICLR 2026 Conference Submission8011 Authors

Keywords: Efficient Reasoning, Attention Outlier, Reasoning

Abstract:

We propose **FROST**, an attention-aware method for efficient reasoning.
Unlike traditional approaches, FROST leverages attention weights to prune uncritical reasoning paths, yielding shorter and more reliable reasoning trajectories.
Methodologically, we introduce the concept of reasoning outliers and design an attention-based mechanism to remove them.
Theoretically, FROST preserves and enhances the model’s reasoning capacity while eliminating outliers at the sentence level.
Empirically, we validate FROST on four benchmarks using two strong reasoning models (Phi-4-Reasoning and GPT-oss-20B), outperforming state-of-the-art methods such as TALE and ThinkLess.
Notably, FROST achieves an average **58.72%** reduction in token usage and a **10.64%** improvement in accuracy over the base model. Furthermore, in evaluations of attention outlier metrics, FROST reduces the maximum infinity norm $\lVert \mathbf{x} \rVert_{\infty}$
by **15.97%** and the average kurtosis by **91.09%** compared to the base model.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 8011