深度思考，信心十足

摘要

大型语言模型（LLMs）在推理任务中展现出巨大潜力，这得益于如自洽多数投票等测试时扩展方法。然而，这种方法往往导致准确率提升有限且计算开销巨大。为应对这些挑战，我们提出了“深度自信思考”（DeepConf），这是一种简单而强大的方法，能在测试时同时提升推理效率和性能。DeepConf利用模型内部置信度信号，在生成过程中或之后动态过滤低质量推理轨迹。该方法无需额外模型训练或超参数调优，并可无缝集成至现有服务框架中。我们在多种推理任务及最新开源模型（包括Qwen 3和GPT-OSS系列）上对DeepConf进行了评估。值得注意的是，在AIME 2025等具有挑战性的基准测试中，DeepConf@512实现了高达99.9%的准确率，并相比全并行思考减少了高达84.7%的生成token数。

English

Large Language Models (LLMs) have shown great potential in reasoning tasks through test-time scaling methods like self-consistency with majority voting. However, this approach often leads to diminishing returns in accuracy and high computational overhead. To address these challenges, we introduce Deep Think with Confidence (DeepConf), a simple yet powerful method that enhances both reasoning efficiency and performance at test time. DeepConf leverages model-internal confidence signals to dynamically filter out low-quality reasoning traces during or after generation. It requires no additional model training or hyperparameter tuning and can be seamlessly integrated into existing serving frameworks. We evaluate DeepConf across a variety of reasoning tasks and the latest open-source models, including Qwen 3 and GPT-OSS series. Notably, on challenging benchmarks such as AIME 2025, DeepConf@512 achieves up to 99.9% accuracy and reduces generated tokens by up to 84.7% compared to full parallel thinking.