Video-KTR: Reinforcing Video Reasoning via Key Token Attribution

Reinforcement learning (RL) has shown strong potential for enhancing reasoning in multimodal large language models, yet existing video reasoning methods often rely on coarse sequence-level rewards or single-factor token selection, neglecting fine-grained links among visual inputs, temporal dynamics, and linguistic outputs, limiting both accuracy and interpretability. We propose Video-KTR, a modality-aware policy shaping framework that performs selective, token-level RL by combining three attribution signals: (1) visual-aware tokens identified via counterfactual masking to reveal perceptual dependence; (2) temporal-aware tokens detected through frame shuffling to expose temporal sensitivity; and (3) high-entropy tokens signaling predictive uncertainty. By reinforcing only these key tokens, Video-KTR focuses learning on semantically informative, modality-sensitive content while filtering out low-value tokens. Across five challenging benchmarks, Video-KTR achieves state-of-the-art or highly competitive results-42.7% on Video-Holmes (surpassing GPT-4o)-with consistent gains on both reasoning and general video understanding tasks. Ablation studies verify the complementary roles of the attribution signals and the robustness of targeted token-level updates. Overall, Video-KTR improves accuracy and interpretability, offering a simple, drop-in extension to RL for complex video reasoning. Our code and models are available at https://github.com/zywang0104/Video-KTR . INTRODUCTION Reinforcement learning (RL) has emerged as a powerful paradigm for enhancing long-chain reasoning in large language models (LLMs) (OpenAI, 2024; DeepSeek-AI, 2025; Yang et al., 2025) . In particular, GRPO (Shao et al., 2024) exhibits strong performance on complex reasoning tasks. Building on this success, RL has been extended to multimodal LLMs (MLLMs), achieving notable gains on visual understanding through sample diversification (