English

Extending Test-Time Scaling: A 3D Perspective with Context, Batch, and Turn

Machine Learning 2025-11-24 v2 Artificial Intelligence

Abstract

Reasoning reinforcement learning (RL) has recently revealed a new scaling effect: test-time scaling. Thinking models such as R1 and o1 improve their reasoning accuracy at test time as the length of the reasoning context increases. However, compared with training-time scaling, test-time scaling is fundamentally limited by the limited context length of base models, which remains orders of magnitude smaller than the amount of tokens consumed during training. We revisit test-time enhancement techniques through the lens of scaling effect and introduce a unified framework of multi-dimensional test-time scaling to extend the capacity of test-time reasoning. Beyond conventional context-length scaling, we consider two additional dimensions: batch scaling, where accuracy improves with parallel sampling, and turn scaling, where iterative self-refinement enhances reasoning quality. Building on this perspective, we propose 3D test-time scaling, which integrates context, batch, and turn scaling. We show that: (1) each dimension demonstrates a test-time scaling effect, but with a bounded capacity; (2) combining all three dimensions substantially improves the reasoning performance of challenging testbeds, including IOI, IMO, and CPHO, and further benefits from human preference feedback; and (3) the human-in-the-loop framework naturally extends to a more open-ended domain, i.e., embodied learning, which enables the design of humanoid control behaviors.

Keywords

Cite

@article{arxiv.2511.15738,
  title  = {Extending Test-Time Scaling: A 3D Perspective with Context, Batch, and Turn},
  author = {Chao Yu and Qixin Tan and Jiaxuan Gao and Shi Yu and Hong Lu and Xinting Yang and Zelai Xu and Yu Wang and Yi Wu and Eugene Vinitsky},
  journal= {arXiv preprint arXiv:2511.15738},
  year   = {2025}
}

Comments

44 pages, 12 figures

R2 v1 2026-07-01T07:45:56.240Z