Settle: Learning When to Stop Reasoning
cs.CL
Submitted: 2026-09-30
Updated: 2026-09-30
Comments: 30 pages, 4 figures
License: http://creativecommons.org/licenses/by/4.0/
The gist: Reasoning models often continue generating after their answers have settled.
Terminology
Abstract
Reasoning models often continue generating after their answers have settled. Settle learns when to stop from answer stability in completed traces. It trains the existing end-of-reasoning token while keeping other predictions close to the base model, and requires only ordinary decoding at inference. On MATH-500 with Qwen3-4B, Settle reduces token count by 40% with a 0.5-percentage-point decrease in accuracy. It gains 6.16 percentage points over supervised fine-tuning on the same traces shortened at their first stable answer, at nearly identical token counts. Its stopping score predicts whether a correct answer will remain correct. Settle extends the accuracy-token-count Pareto frontier of the evaluated stopping methods.
Sources
- LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning
- Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning
- SABER: Stability-Aware Early Exit for LLM Reasoning via Adversarial Branch Probing
- Training Verifiers to Solve Math Word Problems
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems
- Measuring Mathematical Problem Solving With the MATH Dataset
- ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning
- LoRA: Low-Rank Adaptation of Large Language Models
- The Halt Vector: Internalizing a Causal Steering Intervention for Efficient Reasoning
- Efficient Memory Management for Large Language Model Serving with PagedAttention
- Solving Quantitative Reasoning Problems with Language Models
- Decoupled Weight Decay Regularization
- Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
- s1: Simple test-time scaling
- Self-Training Elicits Concise Reasoning in Large Language Models
- TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning
- GPQA: A Graduate-Level Google-Proof Q&A Benchmark
- ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inference
- Qwen3 Technical Report
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering