Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning

2026-08-05

AI Summary

This work addresses the challenge that current large language models struggle to effectively switch between diverse reasoning skills over long-horizon tasks and lack dedicated evaluation and training frameworks for such capabilities. To this end, the study introduces Skill Entropy—a novel metric quantifying the difficulty of cross-skill transitions—and constructs Skill²-Bench, a multi-difficulty benchmark spanning nine domains to evaluate long-range, cross-skill reasoning. Furthermore, Skill Entropy is leveraged as a reward signal in reinforcement learning to design a skill-sequence alignment mechanism that explicitly guides the model to predict the appropriate skill at each reasoning step. Experiments demonstrate substantial improvements: on Qwen3-4B-Instruct and Qwen3-1.7B, the method boosts accuracy on Skill²-Bench from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, significantly outperforming baseline approaches.

Abstract

Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule. We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs. Existing benchmarks often evaluate individual skills, lacking a principled way to measure how well a model switches between skills. We address this gap from both the evaluation and training sides. We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another. We then propose Skill^2-Bench, a benchmark of cross-skill long-horizon tasks built over 558 skills across 9 verifiable and open-ended domains. Each task is assigned a task-level skill-entropy score and grouped into three difficulty levels. Evaluating 8 frontier and 4 open-source models on Skill^2-Bench reveals a skill-switching gap: accuracy decreases on higher-entropy tasks. We then turn skill entropy from a benchmark scale into a training signal. We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it. The reward combines step-level correctness with a skill-entropy reward that measures the alignment between the model-predicted skill sequence and the gold skill sequence. On Qwen3-4B-Instruct and Qwen3-1.7B, Skill-Entropy RL improves the Skill^2-Bench score from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, outperforming competitive baselines. The same pipeline can be applied to off-the-shelf training data such as OpenR1-Math, indicating that skill entropy is a reusable training signal. Code available at: https://github.com/Gen-Verse/Skill-Entropy-RL

Problem

• long-horizon reasoning
• skill switching
• cross-skill tasks
• Skill Entropy
• benchmarking

Innovation

• Skill Entropy
• Long-Horizon Reasoning
• Cross-Skill Tasks
• Reinforcement Learning
• Skill-Switching