Know When to Stop: Segment-Level Credit Assignment for Reducing Overthinking
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Reasoning language models frequently overthink: generating extended chains of behaviors such as hedging, approach abandonment, and self contradiction that consume tokens without improving answers. We show that these behaviors are not merely a consequence of length; even when controlling for response length, incorrect traces exhibit higher rates of unproductive self-reflection than correct ones. Addressing this requires identifying where self-reflection helps vs hurts, but obtaining these step-level annotations is costly. We observe that intermediate answer commitments within reasoning traces can provide a cheap proxy: by comparing each final answer candidate in the trace to the ground truth, we can determine whether subsequent reflection is productive without any additional supervision. Building on this insight, we propose DASH (Drift Aware advantage SHaping), which assigns segment-level credit based on whether each reasoning segment leads toward or away from correctness. On competition-level math benchmarks, DASH achieves the highest accuracy where overthinking is prevalent (Average Accuracy: 59.45% vs. 58.1% Dr.GRPO vs. 56.95% GRPO) while reducing overthinking behaviors and achieving more productive self-correction than baselines.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 17 of its 45 references that are in the map. First posted on arXiv 1 Jul 2026.
Builds on
17 of 45 references.
Closest ancestors · 8
Same direction, newest first
- 2026ROM: Real-time Overthinking Mitigation via Streaming Detection and Intervention23 Mar
- 2025Wait, Wait, Wait... Why Do Reasoning Models Loop?15 Dec
- 2025Revisiting Overthinking in Long Chain-of-Thought from the Perspective of Self-Doubt29 May
- 2025Between Underthinking and Overthinking: An Empirical Study of Reasoning Length and correctness in LLMs30 Apr
- 2025Dynamic Early Exit in Reasoning Models22 Apr
- 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning2 Apr
- 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
- 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
Other references · 9
- 2025VAPO: Efficient and Reliable Reinforcement Learning for Advanced Reasoning Taskscited by 281
- 2025REINFORCE++: Stabilizing Critic-Free Policy Optimization with Global Advantage Normalizationcited by 233
- 2025Phi-4-reasoning Technical Reportcited by 146
- 2025Olmo 3cited by 95
- 2025CAPO: Towards Enhancing LLM Reasoning through Generative Credit Assignmentcited by 25
- 2025GTPO and GRPO-S: Token and Sequence-Level Reward Shaping with Policy Entropycited by 18
- 2025DRPO: Efficient Reasoning via Decoupled Reward Policy Optimizationcited by 16
- 2025λ-GRPO: Unifying the GRPO Frameworks with Learnable Token Preferencescited by 10
- 2026Save the Good Prefix: Precise Error Penalization via Process-Supervised RL to Enhance LLM Reasoningcited by 10
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2026Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
- 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring
- 2025ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding
- 2026REIN: Bridging the Gap between Reasoning and Reliability via Reflection and Abstention Alignment
- 2025The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis
- 2025Lost at the Beginning of Reasoning
- 2026Uncovering the Representation Geometry of Minimal Cores in Overcomplete Reasoning Traces
- 2026Strategic Self-Consistency