Lune

ACL Findings2026

Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggers

Van Yang, Shouren Wang, Debargha Ganguly, Xinpeng Li, Chaoda Song, Vikash Singh, Vipin Chaudhary, Xiaotian Han

Abstract

Hybrid reasoning language models are commonly controlled through high-level Think/Nothink instructions to regulate reasoning behavior, yet we found that such mode switching is largely driven by a small set of trigger tokens rather than the instructions themselves. Through attention analysis and controlled prompting experiments, we show that a leading "Okay" token induces reasoning behavior, while the newline pattern following "</think>" suppresses it. Based on this observation, we propose Mid-Think, a simple training-free prompting format that combines these triggers to achieve intermediate-budget reasoning, consistently outperforming fixedtoken and prompt-based baselines in terms of the accuracy-length trade-off. Furthermore, applying Mid-Think to RL training after SFT reduces training time by approximately 15% while improving final performance of Qwen3-8B on AIME from 69.8% to 72.4% and on GPQA from 58.5% to 61.1%, demonstrating its effectiveness for both inference-time control and RL-based reasoning training. Our code is available at https://github.com/uservan/ Mid-Think.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

1 top-tier paper has cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers1

  1. CausalRAG2: Hierarchical Causal Knowledge Graph Design for RAGNengbo Wang, Tuo Liang, Vikash Singh, Chaoda Song et al.ICML 2026 · 2 citations

Builds on16

  1. DAPO: An Open-Source LLM Reinforcement Learning System at ScaleQiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan et al.NeurIPS 2025 · 2,828 citations
  2. Efficient Streaming Language Models with Attention SinksGuangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han et al.ICLR 2024 · 1,714 citations
  3. Thinkless: LLM Learns When to ThinkGongfan Fang, Xinyin Ma, Xinchao WangNeurIPS 2025 · 128 citations
  4. Think Only When You Need with Large Hybrid-Reasoning ModelsLingjie Jiang, Xun Wu, Shaohan Huang, Qingxiu Dong et al.NeurIPS 2025 · 71 citations
  5. On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language ModelsCharlie Zhang, Graham Neubig, Xiang YueICML 2026 · 58 citations

Related papers

  1. Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RLSongjun Tu, Jiahao Lin, Qichao Zhang, Xiangyu Tian et al.NeurIPS 2025 · 69 citations
  2. Does Reasoning Improve Seeing? Understanding When Vision-Language Models Benefit from ThinkingJing Bi, Luchuan Song, Dingxin Zhang, Pinxin Liu et al.ICML 2026
  3. SABER: Switchable and Balanced Training for Efficient LLM ReasoningKai Zhao, Yanjun Zhao, Jiaming Song, Shien He et al.AAAI 2026 · 9 citations
  4. Fast Quiet-STaR: Thinking Without Thought TokensWei Huang, Yizhe Xiong, Xin Ye, Zhijie Deng et al.EMNLP Findings 2025
  5. AutoL2S: Auto Long-Short Reasoning for Efficient Large Language ModelsFeng Luo, Yu-Neng Chuang, Guanchu Wang, Hoang Anh Duy Le et al.ACL Findings 2026
Dusk over the sea between two cliffs drawn in fine vertical lines