{"ID":2846461,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.01170","arxiv_id":"2511.01170","title":"DART: Difficulty-Adaptive Reasoning Truncation for Efficient Large Language Models","abstract":"Adaptive reasoning is essential for aligning the computational effort of large language models (LLMs) with the intrinsic difficulty of problems. Current chain-of-thought methods boost reasoning ability but indiscriminately generate long explanations, leading to evident inefficiency. However, existing reinforcement learning approaches to adaptive thinking remain unstable and heavily reward-dependent. Here we propose \\textbf{DART}, a supervised \\textbf{D}ifficulty-\\textbf{A}daptive \\textbf{R}easoning \\textbf{T}runcation framework that adjusts thinking length according to problem difficulty. By distilling concise reasoning patterns from stronger models, interpolating them into a continuum of reasoning styles, and curating optimal training data that balances correctness and compactness, DART learns when to ``stop thinking''. Across multiple mathematical benchmarks, experimental results demonstrate its remarkable efficiency while preserving or improving accuracy, achieving a significant 81.2\\% reasoning truncation (DeepSeek-R1-Distill-Qwen-7B on GSM8K dataset) with 5.33$\\times$ computational acceleration. DART provides a stable and general paradigm for efficient reasoning, advancing the development of adaptive intelligence in LLMs.","short_abstract":"Adaptive reasoning is essential for aligning the computational effort of large language models (LLMs) with the intrinsic difficulty of problems. Current chain-of-thought methods boost reasoning ability but indiscriminately generate long explanations, leading to evident inefficiency. However, existing reinforcement lear...","url_abs":"https://arxiv.org/abs/2511.01170","url_pdf":"https://arxiv.org/pdf/2511.01170v2","authors":"[\"Ruofan Zhang\",\"Bin Xia\",\"Zhen Cheng\",\"Cairen Jian\",\"Minglun Yang\",\"Ngai Wong\",\"Yuan Cheng\"]","published":"2025-11-03T02:41:20Z","proceeding":"cs.AI","tasks":"[\"cs.AI\"]","methods":"[\"Reinforcement Learning\",\"Large Language Model\",\"Language Model\"]","has_code":false}