EvoThink: Evolving Thinking in Large Reasoning Models via Self-Pruning and Aha-Moment Preference Optimization
Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 5693-5701.
https://doi.org/10.24963/ijcai.2026/634
Keywords
- Natural Language Processing: Language models
- Natural Language Processing: Applications
Abstract
Large Reasoning Models (LRMs) often suffer from overthinking due to redundant verification steps. Existing approaches for mitigating overthinking, such as fast-slow thinking switching and reasoning trajectory compression, fail to make a fine-grained distinction between beneficial and redundant steps within the LRM's reasoning process, and may thus impair reasoning capability in their pursuit of efficiency. To simultaneously improve reasoning efficiency and capability, we propose EvoThink, a framework that reduces redundant verification and encourages the exploration of new reasoning paths. EvoThink comprises two key components: Self-Pruning Training (SPT), an unsupervised method that iteratively prunes redundant reasoning steps and self-trains on the concise trajectories; and Aha-Moment Preference Optimization (AMPO), which, inspired by genetic algorithms, identifies valuable failed reasoning attempts, synthesizes from-wrong-to-right aha-moment data, and optimizes the model to internalize this reasoning pattern. Extensive evaluations across mathematical reasoning and code generation benchmarks demonstrate that EvoThink not only substantially reduces inference-time token usage but also improves the reasoning capability of LRMs.
@inproceedings{ijcai2026p634,
title = {EvoThink: Evolving Thinking in Large Reasoning Models via Self-Pruning and Aha-Moment Preference Optimization},
author = {Dai, Xinbang and Xin, Zheyu and Hu, Huikang and Ren, Lin and Jin, Rihui and Xiao, Guohui and Dong, Kuicai and Du, Zhaocheng and Zhang, Yuyang},
booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
Artificial Intelligence, {IJCAI-26}},
publisher = {International Joint Conferences on Artificial Intelligence Organization},
editor = {Diego Calvanese},
pages = {5693--5701},
year = {2026},
month = {8},
note = {Main Track},
doi = {10.24963/ijcai.2026/634},
url = {https://doi.org/10.24963/ijcai.2026/634},
}
TY - CPAPER TI - EvoThink: Evolving Thinking in Large Reasoning Models via Self-Pruning and Aha-Moment Preference Optimization AU - Dai, Xinbang AU - Xin, Zheyu AU - Hu, Huikang AU - Ren, Lin AU - Jin, Rihui AU - Xiao, Guohui AU - Dong, Kuicai AU - Du, Zhaocheng AU - Zhang, Yuyang T2 - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26 PB - International Joint Conferences on Artificial Intelligence Organization PY - 2026 SP - 5693 EP - 5701 DO - 10.24963/ijcai.2026/634 UR - https://doi.org/10.24963/ijcai.2026/634 ER -Copied
Machine-readable versions of this paper, for tools and AI agents:
