Test-time Recursive Thinking: Self-Improvement without External Feedback

TL;DR

This paper introduces Test-time Recursive Thinking (TRT), a self-improvement method enabling large language models to enhance reasoning accuracy without external training, achieving significant performance gains on challenging benchmarks.

Contribution

The paper presents TRT, a novel iterative self-improvement framework for LLMs that improves reasoning accuracy without additional training or external feedback.

Findings

Open-source models reach 100% accuracy on AIME-25/24.

Closed-source models improve by 10.4-14.8 percentage points on difficult problems.

TRT enables significant performance gains without external supervision.

Abstract

Modern Large Language Models (LLMs) have shown rapid improvements in reasoning capabilities, driven largely by reinforcement learning (RL) with verifiable rewards. Here, we ask whether these LLMs can self-improve without the need for additional training. We identify two core challenges for such systems: (i) efficiently generating diverse, high-quality candidate solutions, and (ii) reliably selecting correct answers in the absence of ground-truth supervision. To address these challenges, we propose Test-time Recursive Thinking (TRT), an iterative self-improvement framework that conditions generation on rollout-specific strategies, accumulated knowledge, and self-generated verification signals. Using TRT, open-source models reach 100% accuracy on AIME-25/24, and on LiveCodeBench's most difficult problems, closed-source models improve by 10.4-14.8 percentage points without external feedback.