Early Stopping in Large Reasoning Models: A Survey and Empirical Audit of Efficiency and Reliability
cs.AI
Submitted: 2026-10-30
Updated: 2026-11-12
Terminology
Sources
- Activation Steering for Chain-of-Thought Compression
- Probing the Trajectories of Reasoning Traces in Large Language Models
- Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
- Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
- Training Verifiers to Solve Math Word Problems
- S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping
- Early Stopping for Large Reasoning Models via Confidence Dynamics
- Efficient Reasoning via Constrained Optimization in Latent Space
- LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code
- FlashThink: An Early Exit Method For Efficient Reasoning
- Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs
- HaluEval: A Large-Scale Hallucination Evaluation Benchmark for Large Language Models
- Let's Verify Step by Step
- TruthfulQA: Measuring How Models Mimic Human Falsehoods
- Answer Convergence as a Signal for Early Stopping in Reasoning
- Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
- gpt-oss-120b & gpt-oss-20b Model Card
- OpenAI o1 System Card
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents