UniT: Unified Multimodal Chain-of-Thought Test-time Scaling
Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu · 2026
Read UniT: Unified Multimodal Chain-of-Thought Test-time Scaling by Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-12279.txt.
Description
Unified models can both read and make images, yet they usually run in one pass without checking or fixing their own work. Leon Liangyu Chen led this work with 13 co-authors, presenting UniT, a frame for multimodal chain-of-thought test-time scaling. One unified model reasons, checks, and refines across several rounds, handling spatial layouts, many objects, and changing instructions.
UniT joins agentic data synthesis, unified training, and flexible test-time inference to spark checking, subgoal splitting, and content memory. Three findings stand out: short reasoning training generalises to longer test-time chains, sequential reasoning scales better than parallel sampling, and training on generation and editing lifts out-of-distribution visual reasoning. Test-time scaling thus becomes a real lever for unified models.
Catalog record
| CREATOR | Leon Liangyu Chen, Haoyu Ma, Zhipeng Fan, Ziqi Huang, Animesh Sinha, Xiaoliang Dai, Jialiang Wang, Zecheng He, Jianwei Yang, Chunyuan Li, Junzhe Sun, Chu Wang, Serena Yeung-Levy, Felix Juefei-Xu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-12279