The cold-start initialization stage plays a pivotal role in training Multimodal Large Reasoning Models (MLRMs), yet its mechanisms remain insufficiently understood. To analyze this stage, we introduce the Visual Attention Score (VAS), an attention-based metric that quantifies how much a model attends to visual tokens. We find that reasoning performance is strongly correlated with VAS (r=0.9616): models with higher VAS achieve substantially stronger multimodal reasoning. Surprisingly, multimodal cold-start fails to elevate VAS, resulting in attention distributions close to the base model, whereas text-only cold-start leads to a clear increase. We term this counter-intuitive phenomenon Lazy Attention Localization. To validate its causal role, we design training-free interventions that directly modulate attention allocation during inference, performance gains of 1−2% without any retraining. Building on these insights, we further propose Attention-Guided Visual Anchoring and Reflection (AVAR), a comprehensive cold-start framework that integrates visual-anchored data synthesis, attention-guided objectives, and visual-anchored reward shaping. Applied to Qwen2.5-VL-7B, AVAR achieves an average gain of 7.0% across 7 multimodal reasoning benchmarks. Ablation studies further confirm that each component of AVAR contributes step-wise to the overall gains. The code, data, and models are available at https://github.com/lrlbbzl/Qwen-AVAR.
@article{arxiv.2603.03825,
title = {From Narrow to Panoramic Vision: Attention-Guided Cold-Start Reshapes Multimodal Reasoning},
author = {Ruilin Luo and Chufan Shi and Yizhen Zhang and Cheng Yang and Songtao Jiang and Tongkun Guan and Ruizhe Chen and Ruihang Chu and Peng Wang and Mingkun Yang and Yujiu Yang and Junyang Lin and Zhibo Yang},
journal= {arXiv preprint arXiv:2603.03825},
year = {2026}
}