# Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models arXiv: https://arxiv.org/abs/2605.17672 ยท HTML: https://mlsanity.com/paper/2605.17672/ ## Key facts - Cited by 7 papers in the map (data through 14 Oct 2026). - Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. - Builds on 28 of its 91 references that are in the map. - First posted on arXiv 17 May 2026. ## Placed in - LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/) - Efficient LLM inference and serving / [Efficient reasoning and agent inference](https://mlsanity.com/inference/efficient-reasoning-and-agent-inference/) ## Builds on (most cited first, 28 of 28) - [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021) - [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024) - [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025) - [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018) - [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025) - [Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models](https://mlsanity.com/paper/2506.05176/) (2025) - [OpenAI o1 System Card](https://mlsanity.com/paper/2412.16720/) (2024) - [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://mlsanity.com/paper/2501.12599/) (2025) - [Kimi K2: Open Agentic Intelligence](https://mlsanity.com/paper/2507.20534/) (2025) - [GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models](https://mlsanity.com/paper/2508.06471/) (2025) - [SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuning](https://mlsanity.com/paper/2408.05517/) (2024) - [Chain of Draft: Thinking Faster by Writing Less](https://mlsanity.com/paper/2502.18600/) (2025) - [Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety](https://mlsanity.com/paper/2507.11473/) (2025) - [Reasoning Models Can Be Effective Without Thinking](https://mlsanity.com/paper/2504.09858/) (2025) - [Llama-Nemotron: Efficient Reasoning Models](https://mlsanity.com/paper/2505.00949/) (2025) - [Between Underthinking and Overthinking: An Empirical Study of Reasoning Length and correctness in LLMs](https://mlsanity.com/paper/2505.00127/) (2025) - [Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost](https://mlsanity.com/paper/2407.19825/) (2024) - [LLM-Based Human-Agent Collaboration and Interaction Systems: A Survey](https://mlsanity.com/paper/2505.00753/) (2025) - [Entropy After for reasoning model early exiting](https://mlsanity.com/paper/2509.26522/) (2025) - [Group Think: Multiple Concurrent Reasoning Agents Collaborating at Token Level Granularity](https://mlsanity.com/paper/2505.11107/) (2025) - [Early Stopping Chain-of-thoughts in Large Language Models](https://mlsanity.com/paper/2509.14004/) (2025) - [The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis](https://mlsanity.com/paper/2508.17627/) (2025) - [TL;DR: Too Long, Do Re-weighting for Efficient LLM Reasoning Compression](https://mlsanity.com/paper/2506.02678/) (2025) - [Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning](https://mlsanity.com/paper/2510.10103/) (2025) - [MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs](https://mlsanity.com/paper/2511.14159/) (2025) - [PhGPO: Pheromone-Guided Policy Optimization for Long-Horizon Tool Planning](https://mlsanity.com/paper/2602.13691/) (2026) - [LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning](https://mlsanity.com/paper/2512.05325/) (2025) - [NEAT: Neuron-Based Early Exit for Large Reasoning Models](https://mlsanity.com/paper/2602.02010/) (2026) ## Built on by (most cited first, 7 of 7) - [When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models](https://mlsanity.com/paper/2606.30852/) (2026) - [TSRouter: Dynamic Modality-Model Selection for Time Series Reasoning](https://mlsanity.com/paper/2607.08940/) (2026) - [CUSUM-Shaped Inference-Time Monitoring and Targeted Re-Decoding for Quantized Small Language Model Reasoning](https://mlsanity.com/paper/2607.20129/) (2026) - [Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal](https://mlsanity.com/paper/2609.09989/) (2026) - [Settle: Learning When to Stop Reasoning](https://mlsanity.com/paper/2609.38997/) (2026) - [Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use Agents](https://mlsanity.com/paper/2610.02330/) (2026) - [CIPO: Counterfactual Imagination Policy Optimization for Adaptive Tool Granularity Selection](https://mlsanity.com/paper/2610.04991/) (2026) ## Abstract (arXiv) Large Reasoning Models (LRMs) achieve strong performance by generating long chains of thought (CoT), but often overthink, continuing to reason after a solution has already stabilized and thereby wasting tokens and increasing latency. Existing inference-time early-exit methods rely primarily on answer-level signals, such as confidence or trial-answer consistency, to decide when to stop. However, these signals mainly reflect answer readiness rather than reasoning convergence: they may trigger before the model has finished exploring or self-correcting, causing premature exits that can degrade final-answer accuracy and leave the retained reasoning chain semantically incomplete. We identify reasoning-level semantic redundancy as a complementary signal for semantic-preserving early exit: when successive steps no longer add novel progress and instead revisit established conclusions, the reasoning trajectory has likely converged. Building on this insight, we propose PUMA, a plug-and-play framework that combines a lightweight Redundancy Detector with answer-level verification. The detector flags semantically redundant candidate exits, while verification confirms whether stopping is safe, allowing PUMA to remove redundant continuation while preserving both answer accuracy and a coherent reasoning prefix. Across five LRMs and five challenging reasoning benchmarks, PUMA achieves 26.2% average token reduction while preserving accuracy and retained CoT quality. Additional experiments on code generation, zero-shot vision-language reasoning, and learned stopping-policy internalization further demonstrate that reasoning-level redundancy is a robust, transferable, and learnable signal for efficient reasoning. Our code is available at urlhttps://github.com/giovanni-vaccarino/PUMA.