Thinking with Reasoning Skills: Fewer Tokens, More Accuracy

2026-04-23

AI Summary

This work proposes a novel reasoning paradigm that overcomes the inefficiency and high computational cost of large language models (LLMs) caused by generating excessively verbose intermediate steps. Departing from conventional “reasoning from scratch” approaches, the method introduces a retrievable repository of reasoning skills for the first time. Effective reasoning paths discovered through trial-and-error exploration are distilled into a skill library, which the model dynamically queries during inference to guide its reasoning process, thereby avoiding redundant derivations. Evaluated on code generation and mathematical reasoning tasks, this approach significantly reduces token consumption while simultaneously improving accuracy and overall performance, demonstrating both computational efficiency and practical value for real-world deployment.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

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 \emph{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.

Problem

• reasoning tokens
• chain-of-thought
• large language models
• inference efficiency
• computational cost

Innovation

• reasoning skills
• token efficiency
• retrieval-augmented reasoning
• skill distillation
• cost-effective inference