Revisiting the Necessity of Lengthy Chain-of-Thought in Vision-centric Reasoning Generalization

TL;DR

This paper investigates how different Chain-of-Thought (CoT) formats influence the generalization of visual reasoning in vision-language models, revealing that concise CoT with minimal grounding outperforms longer, more detailed approaches.

Contribution

It systematically compares various CoT designs in a controlled maze-solving benchmark, providing new insights into effective CoT strategies for visual reasoning.

Findings

Concise CoT with essential grounding steps outperforms longer traces.

Visual and longer CoT accelerate convergence but do not improve final performance.

Minimal grounding results in better generalization across maze sizes.

Abstract

We study how different Chain-of-Thought (CoT) designs affect the acquisition of the generalizable visual reasoning ability in vision-language models (VLMs). While CoT data, especially long or visual CoT such as "think with image", has been widely used to supervise intermediate reasoning, it remains unclear why specific CoT designs help and which ones truly support generalizable reasoning. To systematically evaluate this, we focus on a controlled maze-solving benchmark where reasoning rules are fully visual, difficulty can be tuned by grid size, and all the intermediate steps can be automatically generated. Using Qwen2.5-VL-7B under a standard SFT-then-RL pipeline, we compare three representative CoT formats: Language CoT, Grounding CoT (with spatial coordinate trajectories), and Visual CoT (with image manipulations). Our experiments reveal that visual and longer CoT mainly accelerate convergence but do not lift the final performance ceiling; concise CoT containing only essential grounding steps outperforms longer traces; and, strikingly, CoT retaining only the minimal grounding results generalizes best across different maze sizes. We further validate these insights on other vision-centric tasks. These findings highlight a "short is long" effect and provide practical guidance for constructing more generalizable SFT datasets for visual reasoning.