While Large Language Models (LLMs) have demonstrated impressive general capabilities, their direct application in the legal domain is often hindered by a lack of precise domain knowledge and complexity of performing rigorous multi-step judicial reasoning. To address this gap, we present LegalOne, a family of foundational models specifically tailored for the Chinese legal domain. LegalOne is developed through a comprehensive three-phase pipeline designed to master legal reasoning. First, during mid-training phase, we propose Plasticity-Adjusted Sampling (PAS) to address the challenge of domain adaptation. This perplexity-based scheduler strikes a balance between the acquisition of new knowledge and the retention of original capabilities, effectively establishing a robust legal foundation. Second, during supervised fine-tuning, we employ Legal Agentic CoT Distillation (LEAD) to distill explicit reasoning from raw legal texts. Unlike naive distillation, LEAD utilizes an agentic workflow to convert complex judicial processes into structured reasoning trajectories, thereby enforcing factual grounding and logical rigor. Finally, we implement a Curriculum Reinforcement Learning (RL) strategy. Through a progressive reinforcement process spanning memorization, understanding, and reasoning, LegalOne evolves from simple pattern matching to autonomous and reliable legal reasoning. Experimental results demonstrate that LegalOne achieves state-of-the-art performance across a wide range of legal tasks, surpassing general-purpose LLMs with vastly larger parameter counts through enhanced knowledge density and efficiency.

Long-video understanding (LVU) is increasingly framed as an agentic process that iteratively localizes task-critical moments before perception and reasoning. Moment localization is therefore a key bottleneck: under long-video context constraints, sparse sampling can easily miss decisive evidence, while similarity-based retrieval often returns semantically related but task-irrelevant clips. We present EviScan, an evidence-seeking moment localization model built on an MLLM, designed for task-oriented localization and near-second temporal scanning on hour-long videos. To learn task-oriented localization, we construct ToMo-Pile, a training set that pairs temporal intervals with rationale traces explaining why the interval resolves the task. To make near-second scanning feasible on long videos, EviScan uses DualKV Prefill for memory-efficient streaming inference, and Progressive Localization to first narrow down candidate intervals and then refine boundaries using high-resolution evidence. We further introduce ToMo-Bench, a benchmark for task-oriented moment localization across diverse long-video scenarios. EviScan consistently outperforms specialized localizers and general-purpose MLLMs on both ToMo-Bench and standard temporal grounding benchmarks. When used as a plug-in localization module, it yields substantial gains on downstream long-video understanding benchmarks (MLVU and LongVideoBench), and these gains are larger than those brought by other moment-localization methods. We will release the model, dataset, benchmark, and code.

Pre-trained language models like BERT and T5 serve as crucial backbone encoders for dense retrieval. However, these models often exhibit limited generalization capabilities and face challenges in improving in-domain accuracy. Recent research has explored using large language models (LLMs) as retrievers, achieving state-of-the-art performance across various tasks. Despite these advancements, the specific benefits of LLMs over traditional retrievers and the impact of different LLM configurations—such as parameter sizes, pre-training duration, and alignment processes—on retrieval tasks remain unclear. In this work, we conduct a comprehensive empirical study on a wide range of retrieval tasks, including in-domain accuracy, data efficiency, zero-shot generalization, lengthy retrieval, instruction-based retrieval, and multi-task learning. We evaluate over 15 different backbone LLMs and non-LLMs. Our findings reveal that larger models and extensive pre-training consistently enhance in-domain accuracy and data efficiency. Additionally, larger models demonstrate significant potential in zero-shot generalization, lengthy retrieval, instruction-based retrieval, and multi-task learning. These results underscore the advantages of LLMs as versatile and effective backbone encoders in dense retrieval, providing valuable insights for future research and development in this field.