MatryoshkaThinking: Recursive Test-Time Scaling Enables Efficient Reasoning

2025-10-11

AI Summary

Test-time scaling often incurs prohibitive computational costs, making it difficult to balance performance and efficiency. Method: This paper proposes a recursive reasoning-verification-summarization framework that iteratively leverages the model’s own capabilities to refine and correct multi-step reasoning processes. Contribution/Results: The method significantly reduces computational overhead while maintaining or even improving solution accuracy: it achieves 99.79 on AIME2025 using only 4% of DeepConf’s FLOPs. It substantially narrows the gap between Pass@k and Pass@1, enhancing reasoning consistency and robustness. The framework demonstrates strong generalizability—validated across multiple open-source language models and multimodal benchmarks. By enabling efficient, self-correcting inference, it establishes a novel paradigm for scalable, reliable test-time adaptation.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Test-time scaling has emerged as a promising paradigm in language modeling, wherein additional computational resources are allocated during inference to enhance model performance. Recent approaches, such as DeepConf, have demonstrated the efficacy of this strategy, however, they often incur substantial computational overhead to achieve competitive results. In this work, we propose MatryoshkaThinking, a novel method that significantly reduces computational cost while maintaining state-of-the-art performance. Specifically, MatryoshkaThinking attains a score of 99.79 on AIME2025 using only 4% of the computation required by DeepConf. The core of our approach lies in the recursive exploitation of the model's intrinsic capabilities in reasoning, verification, and summarization, which collectively enhance the retention of correct solutions and reduce the disparity between Pass@k and Pass@1. Comprehensive evaluations across multiple open-source models and challenging multi-modal reasoning benchmarks validate the effectiveness and generality of our method. These findings offer new insights into the design of efficient and scalable test-time inference strategies for advanced language models.

Problem

• Reduces computational cost while maintaining top performance
• Recursively exploits reasoning, verification, and summarization capabilities
• Enhances solution retention and reduces Pass@k vs Pass@1 disparity

Innovation

• Recursive test-time scaling reduces computational costs
• Exploits model's reasoning, verification, and summarization capabilities
• Enhances solution retention and reduces Pass@k disparity