用模型内部置信度动态过滤低质量推理，提升效率与准确率。
Deep Think with Confidence
- 基于模型自身置信度筛选推理路径，动态剔除无效生成。
- 在AIME 2025上达99.9%准确率，生成词数减少84.7%。
- 无需训练或调参，可无缝接入现有推理系统。
大语言模型在测试时通过自一致性等方法实现推理能力提升，但常面临准确率增长放缓和计算开销高的问题。为此，我们提出深度思考置信度机制（DeepConf），利用模型内部置信信号，在生成过程中或之后动态过滤低质量推理路径。该方法无需额外训练或超参数调优，可直接集成到现有服务框架中。我们在多种推理任务及最新开源模型（如Qwen 3、GPT-OSS系列）上进行了评估。尤其在挑战性基准AIME 2025上，DeepConf@512达到最高99.9%准确率，相比完整并行思考，生成词数减少高达84.7%。
原文摘要 · Abstract (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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。