# Entropy After for reasoning model early exiting arXiv: https://arxiv.org/abs/2509.26522 ยท HTML: https://mlsanity.com/paper/2509.26522/ ## Key facts - Cited by 24 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 32 of its 46 references that are in the map. - First posted on arXiv 30 Sep 2025. ## 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, 30 of 32) - [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022) - [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023) - [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023) - [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023) - [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024) - [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024) - [s1: Simple test-time scaling](https://mlsanity.com/paper/2501.19393/) (2025) - [BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks](https://mlsanity.com/paper/1709.01686/) (2017) - [Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs](https://mlsanity.com/paper/2412.21187/) (2024) - [Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning](https://mlsanity.com/paper/2506.01939/) (2025) - [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022) - [Adaptive Computation Time for Recurrent Neural Networks](https://mlsanity.com/paper/1603.08983/) (2016) - [Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models](https://mlsanity.com/paper/2503.16419/) (2025) - [Confident Adaptive Language Modeling](https://mlsanity.com/paper/2207.07061/) (2022) - [Reasoning with Exploration: An Entropy Perspective](https://mlsanity.com/paper/2506.14758/) (2025) - [Depth-Adaptive Transformer](https://mlsanity.com/paper/1910.10073/) (2019) - [Dynamic Early Exit in Reasoning Models](https://mlsanity.com/paper/2504.15895/) (2025) - [Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification](https://mlsanity.com/paper/2504.05419/) (2025) - [Rank-normalization, folding, and localization: An improved widehatR for assessing convergence of MCMC](https://mlsanity.com/paper/1903.08008/) (2019) - [S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models](https://mlsanity.com/paper/2505.07686/) (2025) - [Jointly Reinforcing Diversity and Quality in Language Model Generations](https://mlsanity.com/paper/2509.02534/) (2025) - [Sample More to Think Less: Group Filtered Policy Optimization for Concise Reasoning](https://mlsanity.com/paper/2508.09726/) (2025) - [Answer Matching Outperforms Multiple Choice for Language Model Evaluation](https://mlsanity.com/paper/2507.02856/) (2025) - [Answer Convergence as a Signal for Early Stopping in Reasoning](https://mlsanity.com/paper/2506.02536/) (2025) - [Understanding Chain-of-Thought in LLMs through Information Theory](https://mlsanity.com/paper/2411.11984/) (2024) - [Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lens](https://mlsanity.com/paper/2505.18237/) (2025) - [Fast yet Safe: Early-Exiting with Risk Control](https://mlsanity.com/paper/2405.20915/) (2024) - [Optimizing Anytime Reasoning via Budget Relative Policy Optimization](https://mlsanity.com/paper/2505.13438/) (2025) - [LLM Probability Concentration: How Alignment Shrinks the Generative Horizon](https://mlsanity.com/paper/2506.17871/) (2025) - [RLP: Reinforcement as a Pretraining Objective](https://mlsanity.com/paper/2510.01265/) (2025) ## Built on by (most cited first, 24 of 24) - [Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing](https://mlsanity.com/paper/2602.03845/) (2026) - [Are Your Reasoning Models Reasoning or Guessing? A Mechanistic Analysis of Hierarchical Reasoning Models](https://mlsanity.com/paper/2601.10679/) (2026) - [Conformal Thinking: Risk Control for Reasoning on a Compute Budget](https://mlsanity.com/paper/2602.03814/) (2026) - [LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling](https://mlsanity.com/paper/2605.08083/) (2026) - [Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models](https://mlsanity.com/paper/2605.17672/) (2026) - [TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning](https://mlsanity.com/paper/2603.12529/) (2026) - [Early Stopping for Large Reasoning Models via Confidence Dynamics](https://mlsanity.com/paper/2604.04930/) (2026) - [PRInTS: Reward Modeling for Long-Horizon Information Seeking](https://mlsanity.com/paper/2511.19314/) (2025) - [On-Policy Supervised Fine-Tuning for Efficient Reasoning](https://mlsanity.com/paper/2602.13407/) (2026) - [Unveiling the Entropy Dynamics of Chain-of-Thought Reasoning](https://mlsanity.com/paper/2606.02020/) (2026) - [Controlling the Risk of Corrupted Contexts for Language Models via Early-Exiting](https://mlsanity.com/paper/2510.02480/) (2025) - [interwhen: A Generalizable Framework for Steering Reasoning Models with Test-time Verification](https://mlsanity.com/paper/2602.11202/) (2026) - [The Potential of CoT for Reasoning: A Closer Look at Trace Dynamics](https://mlsanity.com/paper/2602.14903/) (2026) - [When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models](https://mlsanity.com/paper/2606.30852/) (2026) - [Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck](https://mlsanity.com/paper/2603.08462/) (2026) - [ROM: Real-time Overthinking Mitigation via Streaming Detection and Intervention](https://mlsanity.com/paper/2603.22016/) (2026) - [Think Twice Before You Write -- an Entropy-based Decoding Strategy to Enhance LLM Reasoning](https://mlsanity.com/paper/2604.00018/) (2026) - [DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?](https://mlsanity.com/paper/2606.12402/) (2026) - [SABER: Stability-Aware Early Exit for LLM Reasoning via Adversarial Branch Probing](https://mlsanity.com/paper/2608.27963/) (2026) - [LiteStage: Latency-aware Layer Skipping for Multi-stage Reasoning](https://mlsanity.com/paper/2510.14211/) (2025) - [Step-Tagging: Toward controlling the generation of Language Reasoning Models through step monitoring](https://mlsanity.com/paper/2512.14332/) (2025) - [TRACES: Tagging Reasoning Steps for Adaptive Cost-Efficient Early-Stopping](https://mlsanity.com/paper/2604.21057/) (2026) - [Rewarding Efficient Reasoning Improves Abstention on Underspecified Tasks in Reasoning Models](https://mlsanity.com/paper/2609.20846/) (2026) - [Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency](https://mlsanity.com/paper/2609.31619/) (2026) ## Abstract (arXiv) Reasoning LLMs show improved performance with longer chains of thought. However, recent work has highlighted their tendency to overthink, continuing to revise answers even after reaching the correct solution. We quantitatively confirm this inefficiency from the distribution dynamics perspective by tracking Pass@1 for answers averaged over a large number of rollouts and find the model often begins to always produce the correct answer early in the reasoning, making extra reasoning tokens wasteful. To detect and prevent overthinking, we propose a simple and inexpensive novel signal, Entropy After (EAT), for monitoring and deciding whether to exit reasoning early. By appending a stop thinking token () and monitoring the entropy of the following token as the model reasons, we obtain a trajectory that decreases and stabilizes when Pass@1 plateaus; thresholding its variance under an exponential moving average yields a practical stopping rule. Importantly, our approach enables adaptively allocating compute based on the EAT trajectory, allowing us to spend compute in a more efficient way compared with fixing the token budget for all questions. Empirically, on MATH500 and AIME2025, EAT reduces token usage by 12 - 22% without harming accuracy. EAT also remains effective in black box settings where logits from the reasoning model are not accessible, and EAT is computed with proxy models: We verified the feasibility via early stopping Llama 70B with a 1.5B model and Claude 3.7 with a local 4B model.