Summary
Multimodal AI models that understand pictures and text usually explain their thinking using long natural language sentences, which can lead to confusion and less accurate answers. The authors propose a new way called Dense Reasoning Trace (DRT) that uses short, structured steps instead of full sentences, separating what the model sees from its logical thinking. This approach makes the explanation more efficient and better grounded in the visual information. Their method also uses reinforcement learning to make the model generate these clear, concise reasoning steps with fewer mistakes. Tests show this method is much more efficient and slightly more accurate compared to previous models.
What this means in practice
- •For multimodal ai developers: Create more efficient and accurate image-and-text AI systems by implementing structured, concise reasoning traces instead of verbose natural language explanations.
- •For visual data analytics teams: Improve the reliability and interpretability of automated visual reasoning systems by adopting reasoning traces that reduce hallucinations and strengthen logical grounding.
Authors
Wan Xu, Yuanfan Guo, Kevin Han, LaLa Chen, Wangmeng Zuo
Abstract
Despite the remarkable progress in Multimodal Large Language Models (MLLMs), prevailing Chain-of-Thought (CoT) paradigms remain confined to the natural-language expression space. Consequently, they inherently incur excessive linguistic overhead, leading to information dilution and weak visual grounding. To address this challenge, we propose Dense Reasoning Trace (DRT), a paradigm that departs from natural-language-centered CoT by expressing reasoning as compact structured traces, which include concise intermediate states with symbolic connectors and disentangle visual observations from logical deductions. First, we introduce the Dense Trace Initialization to internalize the DRT reasoning mode into the model, substantially improving token efficiency while preserving visual evidence. To further enable the model to faithfully capture the logical relations within traces, we propose the Trace-Grounded Reinforcement Learning framework, which builds reference traces through a tri-perspective verification pipeline and employs Trace-Grounded GRPO with structured rewards, encouraging the model to generate concise DRT-style traces with reduced hallucination and stronger logical grounding. Extensive experiments on challenging reasoning benchmarks show that DRT achieves 5.5$\times$ token efficiency improvement while improving 1.3 accuracy points over the Qwen3-VL baseline. These findings suggest that complex multimodal reasoning may not require verbose natural-language traces, opening a more efficient path for next-generation MLLMs. Our code and data are available at: https://github.com/HIT-leaderone/DRT