Thinking with Reasoning Skills: Fewer Tokens, More Accuracy

arXiv:2604.21764v1 Announce Type: new Abstract: Reasoning LLMs often spend substantial tokens on long intermediate reasoning traces (e.g., chain-of-thought) when solving new problems. We propose to summarize and store reusable reasoning skills distilled from extensive deliberation and trial-and-error exploration, and to retrieve these skills at inference time to guide future reasoning. Unlike the prevailing reasoning from scratch paradigm, our approach first recalls relevant skills for each query, helping the model avoid redundant detours and focus on effective solution paths. We evaluate our method on coding and mathematical reasoning tasks, and find that it significantly reduces reasoning tokens while improving overall performance. The resulting lower per-request cost indicates strong practical and economic potential for real-world deployment.

How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning
SkillFactory: Self-Distillation For Learning Cognitive Behaviors
Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization
A Mechanistic Analysis of Looped Reasoning Language Models
Sample Complexity of Autoregressive Reasoning: Chain-of-Thought vs. End-to-End

Source: arXiv cs.AI | 2026-04-24

Related