ACL Findings2026
SpiralThinker: Latent Reasoning through an Iterative Process with Text-Latent Interleaving
Abstract
Recent advances in large reasoning models have been driven by reinforcement learning and test-time scaling, accompanied by growing interest in latent rather than purely textual reasoning. However, existing latent reasoning methods lack mechanisms to ensure stable reasoning dynamics in latent space and a systematic way to interleave implicit and explicit reasoning. We introduce SpiralThinker, a stabilized iterative latent reasoning framework that performs iterative updates over latent representations while interleaving latent and textual reasoning steps. At its core, it combines a progressive alignment objective that explicitly regulates latent representations across iterations with structured annotations for text-latent interleaving, thereby stabilizing latent updates and maintaining coherence with textual reasoning. Across mathematical, logical, and commonsense reasoning tasks, SpiralThinker achieves state-of-the-art performance among latent reasoning baselines. Further analysis shows that both iteration and alignment are essential, that the optimal numbers of latent tokens and iterations vary by dataset, and that proper alignment is crucial for effective iterative latent reasoning. Overall, SpiralThinker bridges iterative computation and latent reasoning, demonstrating that aligned iterative updates can reliably steer reasoning in the latent space. Code is available at https://github.com/shengminp/SpiralThinker.
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.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Think before you speak: Training Language Models With Pause Tokens cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models cites Think before you speak: Training Language Models With Pause Tokens
- SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Chain-of-Thought Reasoning Without Prompting cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Cited by top-tier papers1
Builds on12
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu et al.ICLR 2022 · 18,833 citations
- Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth ApproachJonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer et al.NeurIPS 2025 · 431 citations
- Chain-of-Thought Reasoning Without PromptingXuezhi Wang, Denny ZhouNeurIPS 2024 · 305 citations
- Think before you speak: Training Language Models With Pause TokensSachin Goyal, Ziwei Ji, Ankit Singh Rawat, Aditya Krishna Menon et al.ICLR 2024 · 240 citations
Related papers
- Vision-aligned Latent Reasoning for Multi-modal Large Language ModelByungwoo Jeon, Yoonwoo Jeong, Hyunseok Lee, Minsu Cho et al.ICML 2026 · 7 citations
- One Refiner to Unlock Them All: Inference-Time Reasoning Elicitation via Reinforcement Query RefinementYixiao Zhou, Dongzhou Cheng, Zhiliang Wu, Yi Yang et al.ACL 2026 · 3 citations
- ChainGPT: Dual-Reasoning Model with Recurrent Depth and Multi-Rank State UpdatesYunao Zheng, Xiaojie Wang, Lei Ren, Chen WeiICLR 2026
- Internalizing Explicit Reasoning into Latent Space for Dense RetrievalJiajie Jin, Yanzhao Zhang, Mingxin Li, Dingkun Long et al.SIGIR 2026
- Latent-Guided Reasoning: Empowering Small LLMs with Large-Model ThinkingHanzhu Chen, Lin Yang, Jie Wang, Junhao Yan et al.ICLR 2026
