Test-Time Scaling

Follow this topic Ask about this topic

Test-time scaling studies how models can use additional computation after training to solve difficult problems more accurately or reliably. Common approaches include parallel sampling and self-consistency, search and iterative refinement, verifier- or reward-guided selection, and adaptive stopping that allocates effort according to uncertainty or task difficulty. Methods are tested in settings such as mathematical reasoning, code generation, web and tool-using agents, multimodal understanding, and robotics. A central open tension is that more computation can help only when candidate selection and feedback are reliable; correlated errors, weak verifiers, overthinking, and inference cost can erase or reverse the gains.

Latest papers

232 papers32 papers in the last four weeks, up 14% on the four weeks before. 0.2% of all new papers.

Duel-Evolve: Reward-Free Test-Time Scaling via LLM Self-Preferences
Feb 25, 2026Sweta Karlekar, Carolina Zheng, Magnus Saebo +5Large Language Model-Guided Optimization Test-Time Scaling
Columbia University · Los Alamos National Laboratory
From Scale to Speed: Adaptive Test-Time Scaling for Image Editing
Feb 24, 2026Xiangyan Qu, Zhenlong Yuan, Jing Tang +9Test-Time Scaling Text-Guided Image Editing
Institute of Information Engineering, Chinese Academy of Sciences · School of Cyber Security, University of Chinese Academy of Sciences · AMAP, Alibaba Group +1
Limited Reasoning Space: The cage of long-horizon reasoning in LLMs
Feb 22, 2026Zhenyu Li, Guanlin Wu, Cheems Wang +1Long-Horizon LLM Reasoning Long-Horizon Reasoning
Academy of Sciences, Beijing, China · Department of Automation, Tsinghua University, Beijing, China · Department of Computer Science, Peking University, Beijing, China
Evaluating Test-Time Scaling of General LLM Agents
Feb 22, 2026Xiaochuan Li, Ryan Ming, Pranav Setlur +6LLM Agent Evaluation Test-Time Scaling
Language Technologies Institute, School of Computer Science, Carnegie Mellon University · Meta
Team of Thoughts: Efficient Test-time Scaling of Agentic Systems through Orchestrated Tool Calling
Feb 18, 2026Jeffrey T. H. Wong, Zixi Zhang, Junyi Liu +1Multi-Agent LLM Systems Multi-Agent Systems
Imperial College London · Microsoft Research
PETS: A Principled Framework Towards Optimal Trajectory Allocation for Efficient Test-Time Self-Consistency
Feb 18, 2026Zhangyi Liu, Huaizhi Qu, Xiaowei Yin +4Inference-Time Compute Allocation Self-Consistency in Language Models
Stanford University · UNC at Chapel Hill · Yale University +1
Inner Loop Inference for Pretrained Transformers: Unlocking Latent Capabilities Without Training
Feb 16, 2026Jonathan Lys, Vincent Gripon, Bastien Pasdeloup +4Transformer Inference Test-Time Scaling
IMT Atlantique, Lab-STICC, UMR CNRS 6285, Brest, France · Sony Europe Ltd. Stuttgart Technology Center, EUREC, Germany
CodeScaler: Scaling Code LLM Training and Test-Time Inference via Reward Models
Feb 4, 2026Xiao Zhu, Xinyu Zhou, Boyu Zhu +5RL for Code Generation Reward Modeling
LARK, HKUST(GZ) · Kuaishou Technology · UCL +3
ARTIS: Agentic Risk-Aware Test-Time Scaling via Iterative Simulation
Feb 2, 2026Xingshan Zeng, Lingzhi Wang, Weiwen Liu +5LLM Agent Reliability Test-Time Scaling
Huawei Technologies Co., Ltd · Harbin Institute of Technology, Shenzhen · Shanghai Jiao Tong University
What If We Allocate Test-Time Compute Adaptively?
Feb 1, 2026Ahsan Bilal, Ahmed Mohsin, Muhammad Umer +4Inference-Time Search Test-Time Scaling
University of Oklahoma, OK, USA · Stanford University, CA, USA · Universitat Pompeu Fabra, Barcelona, Spain +2
GAIA: A Data Flywheel System for Training GUI Test-Time Scaling Critic Models
Jan 26, 2026Shaokang Wang, Pei Fu, Ruoceng Zhang +7GUI Agents Test-Time Scaling
MiLM Plus, Xiaomi Inc, Beijing, China · School of Computer Science, Peking University, Beijing, China
Asymptotic Universal Alignment: A New Alignment Framework via Test-Time Scaling
Jan 13, 2026Yang Cai, Weiqiang ZhengLLM Alignment Test-Time Scaling
Yale University
CASHEW: Stabilizing Multimodal Reasoning via Iterative Trajectory Aggregation
Jan 12, 2026Chaoyu Li, Fei Tao, Pooyan FazliTest-Time Scaling for VLMs Test-Time Scaling
Arizona State University · NewsBreak
AdaFuse: Adaptive Ensemble Decoding with Test-Time Scaling for LLMs
Jan 9, 2026Chengming Cui, Tianxin Wei, Ziyi Chen +6Language Model Decoding Test-Time Scaling
University of Illinois Urbana-Champaign
Understanding the Effects of Distractors on Reasoning Vision-Language Models
Nov 26, 2025Jiyun Bae, Hyunjong Ok, Sangwoo Mo +1VLM Reasoning Visual Question Answering
Pohang University of Science and Technology (POSTECH)
Refine Thought: A Test-Time Inference Method for Embedding Model Reasoning
Oct 14, 2025Guangzhi Wang, Kai Li, Yinghao Jiao +1Test-Time Scaling Text Embedding Models
CareerInternational Research Team
Representation-Based Exploration for Language Models: From Test-Time to Post-Training
Oct 13, 2025Jens Tuyls, Dylan J. Foster, Akshay Krishnamurthy +1RL for Language Model Reasoning Test-Time Scaling
Princeton University · Microsoft Research NYC
Advancing the Pareto Frontier of Test-Time Scaling with Multi-Path Plan Aggregation
Oct 13, 2025Siheng Xiong, Ali Payani, Faramarz FekriTest-Time Scaling LLM Reasoning
Georgia Institute of Technology · Cisco Research
Verifier-free Test-Time Sampling for Vision-Language-Action Models
Oct 7, 2025Suhyeok Jang, Dongyoung Kim, Changyeon Kim +2Test-Time Scaling Vision-Language-Action Models
1KAIST · 3RLWRLD · 2Seoul National University
Rethinking Reward Models for Multi-Domain Test-Time Scaling
Oct 1, 2025Dong Bok Lee, Seanie Lee, Sangwoo Park +12Reward Modeling LLM Evaluation
KAIST · TH Nürnberg · Microsoft Research Asia
ScalingAR: Scaling Confidence for Autoregressive Image Generation
Sep 30, 2025Harold Haodong Chen, Xianfeng Wu, Wen-Jie Shu +4Language Model Decoding Visual Autoregressive Models
HKUST(GZ) · UNC-Chapel Hill · ZODA +3
Explore-Execute Chain: Towards an Efficient Structured Reasoning Paradigm
Sep 28, 2025Kaisen Yang, Tinghe Zhang, Rushi Shah +4LLM Reasoning Structured Reasoning
1Qizhi Institute · 2Dept. of Computer Science, Tsinghua University · College of AI, Tsinghua University +4

Feb 25, 2026J/K move · Enter open · S save

Duel-Evolve: Reward-Free Test-Time Scaling via LLM Self-Preferences

Sweta Karlekar, Carolina Zheng, Magnus Saebo, Nicolas Beltran-Velez+4

Columbia University · Los Alamos National Laboratory

Large Language Model-Guided Optimization Test-Time Scaling Language Model Self-Improvement Evolutionary Optimization Language Model Self-Assessment

Open paper AskPDF arXiv

Many applications seek to optimize LLM outputs at test time by iteratively proposing, scoring, and refining candidates over a discrete output space. Existing methods use a calibrated scalar evaluator for the target objective to guide search, but for many tasks such scores are unavailable, too sparse, or unreliable. Pairwise comparisons, by contrast, are often easier to elicit, still provide useful signal on improvement directions, and can be obtained from the LLM itself without external supervision. Building on this observation, we introduce Duel-Evolve, an evolutionary optimization algorithm that replaces external scalar rewards with pairwise preferences elicited from the same LLM used to generate candidates. Duel-Evolve aggregates these noisy candidate comparisons via a Bayesian Bradley-Terry model, yielding uncertainty-aware estimates of candidate quality. These quality estimates guide allocation of the comparison budget toward plausible optima using Double Thompson Sampling, as well as selection of high-quality parents to generate improved candidates. We evaluate Duel-Evolve on MathBench, where it achieves 20 percentage points higher accuracy over existing methods and baselines, and on LiveCodeBench, where it improves over comparable iterative methods by over 12 percentage points. Notably, the method requires no reward model, no ground-truth labels during search, and no hand-crafted scoring function. Results show that pairwise self-preferences provide strong optimization signal for test-time improvement over large, discrete output spaces.