AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

SteadyThought: Mitigating LLM Under-Thinking via Thought-Level Preference Optimization

Zhenyue Peng, Lin Yang, Jie Wang, Xiaoqi Ni, Hanzhu Chen, Xize Liang, Yanjun Wu, RujingWang, Kaiwen Zhou, Jianye Hao

NeurIPS 2026 · Large language models › Reasoning and chain of thought · approach: Efficient and adaptive thinking · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes SteadyThought, a thought-level preference optimization method that targets LRM under-thinking (excessive thought switching that abandons promising reasoning trajectories).

Problem

Large Reasoning Models switch between reasoning thoughts excessively and fail to sustain promising ones (under-thinking); prior fixes that suppress switching may discard valuable trajectories.

Approach

Mitigates under-thinking via thought-level preference optimization rather than simply suppressing thought switching, aiming to keep valuable reasoning trajectories while reducing premature switching. Full method details are not available from the abstract.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply Steady Thought preference optimization to the reasoning modelQwen3-8B · 4-benchmark overallaccuracy and tokens: 80.23 -> 84.23 accuracy; tokens 6122 -> 4011Improved
suppress switching tokens with NOWAIT instead of STDeepSeek-R1-Distill-Qwen-14B · 4-benchmark overallaccuracy: 79.73 -> 75.88Hurt
use NoThink baselineall four models · 4-benchmark overallaccuracy: 55.43 -> 37.20 (1.5B)Hurt
apply SEAL steeringQwen3-8B · MuSR / StrategyQA / MedQA-USMLE overallaccuracy: 72.83 -> 35.50Hurt
train only on math, test on LiveCode codingSteady Thought, Qwen3-8B · LiveCodeaccuracy: 71.8 -> 83.2Improved
train on English math, test on Chinese GAOKAO-benchSteady Thought, Qwen3-8B · GAOKAO-benchaccuracy: 86.0 -> 88.4Improved
evaluate from incorrect first-thought prefixes after STSteady Thought, 1.5B · Controlled recovery setaccuracy: 13.2 -> 20.3Improved
use thought-level STPO instead of whole-trajectory SimPO (ablation)Qwen3-8B · AIME 2024accuracy: 60.8 -> 64.6Improved
use STPO instead of DPO at thought level (ablation)DeepSeek-R1-Distill-Qwen-1.5B · MATH-500 tokenstokens: 4428 -> 2691Improved
use SFT on chosen continuations instead of preference optimization (ablation)DeepSeek-R1-Distill-Qwen-1.5B · AIME 2024accuracy: 27.5 -> 20.8 (SFT-thought)Hurt
add explicit linguistic markers to switching-point detection (ablation)Steady Thought · AIME 2024accuracy: 29.2 -> 30.8 (1.5B); 63.3 -> 64.6 (Qwen3-8B)Improved
add boundary entropy spikes to switching-point detection (ablation)Steady Thought · MATH-500 tokenstokens: 3554 -> 2691 (1.5B); 3642 -> 2139 (Qwen3-8B)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy