Yangyang Liu

Kimi Team, Tongtong Bai, Yifan Bai et al.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Yangyang Liu, Dong Yu, Pengyuan Liu

Large language models are increasingly used in value-sensitive decision settings, where irrelevant demographic cues should not alter judgments. We construct the Realistic Value Decision Benchmark (RVDB), a controlled benchmark that varies only the role-gender configuration while holding the scenario, ordered value pair, roles, candidate decisions, Value Distance, and Decision Severity fixed. Using a position-balanced evaluation across seven models, we test whether models preserve decision invariance under gender perturbations and whether their self-attributions reflect observed behavioral changes. We find that explicit gender cues induce bounded but systematic decision flips, including under an explicit gender-attribution prompt that asks models to report whether gender influenced their choice. Cross-gender role swaps reveal a consistent female-proposed-decision asymmetry, while models often attribute flipped decisions to No Influence or other non-gender factors. Further analysis shows that gender effects concentrate near less determinate value boundaries and under more severe decision contexts, suggesting that gender cues act as local boundary-shifting factors rather than global overrides of value reasoning. Value rankings remain largely stable, but ordered value-pair trade-offs shift unevenly across role-gender configurations. These results show that gender can enter LLM value trade-offs behaviorally while remaining obscured in self-attribution, motivating controlled behavioral audits beyond explanation-based evaluation.

KimiTeam, Ding Ding, Zeqian Ju et al.

We present Kimi-Audio, an open-source audio foundation model that excels in audio understanding, generation, and conversation. We detail the practices in building Kimi-Audio, including model architecture, data curation, training recipe, inference deployment, and evaluation. Specifically, we leverage a 12.5Hz audio tokenizer, design a novel LLM-based architecture with continuous features as input and discrete tokens as output, and develop a chunk-wise streaming detokenizer based on flow matching. We curate a pre-training dataset that consists of more than 13 million hours of audio data covering a wide range of modalities including speech, sound, and music, and build a pipeline to construct high-quality and diverse post-training data. Initialized from a pre-trained LLM, Kimi-Audio is continual pre-trained on both audio and text data with several carefully designed tasks, and then fine-tuned to support a diverse of audio-related tasks. Extensive evaluation shows that Kimi-Audio achieves state-of-the-art performance on a range of audio benchmarks including speech recognition, audio understanding, audio question answering, and speech conversation. We release the codes, model checkpoints, as well as the evaluation toolkits in https://github.com/MoonshotAI/Kimi-Audio.

Kimi Team, Yifan Bai, Yiping Bao et al. · tsinghua

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

Ling Shi, Xinwei Wu, Xiaohu Zhao et al.

While mechanistic interpretability tools like Sparse Autoencoders (SAEs) can uncover meaningful features within Large Language Models (LLMs), a critical gap remains in transforming these insights into practical actions for model optimization. We bridge this gap with the hypothesis that data selection guided by a model's internal task features is a effective training strategy. Inspired by this, we propose Interpretability-Guided Data Selection (IGDS), a framework that first identifies these causal task features through frequency recall and interventional filtering, then selects ``Feature-Resonant Data'' that maximally activates task features for fine-tuning. We validate IGDS on mathematical reasoning, summarization, and translation tasks within Gemma-2, LLaMA-3.1, and Qwen3 models. Our experiments demonstrate exceptional data efficiency: on the Math task, IGDS surpasses full-dataset fine-tuning by a remarkable 17.4% on Gemma-2-2B while using only 50% of the data, and outperforms established baselines focused on data quality and diversity. Analysis confirms a strong positive correlation between feature amplification and task performance improvement. IGDS thus provides a direct and effective framework to enhance LLMs by leveraging their internal mechanisms, validating our core hypothesis.

Yipeng Leng, Qiangjuan Huang, Zhiyuan Wang et al.

Diffusion probabilistic models (DPMs) have shown remarkable results on various image synthesis tasks such as text-to-image generation and image inpainting. However, compared to other generative methods like VAEs and GANs, DPMs lack a low-dimensional, interpretable, and well-decoupled latent code. Recently, diffusion autoencoders (Diff-AE) were proposed to explore the potential of DPMs for representation learning via autoencoding. Diff-AE provides an accessible latent space that exhibits remarkable interpretability, allowing us to manipulate image attributes based on latent codes from the space. However, previous works are not generic as they only operated on a few limited attributes. To further explore the latent space of Diff-AE and achieve a generic editing pipeline, we proposed a module called Group-supervised AutoEncoder(dubbed GAE) for Diff-AE to achieve better disentanglement on the latent code. Our proposed GAE has trained via an attribute-swap strategy to acquire the latent codes for multi-attribute image manipulation based on examples. We empirically demonstrate that our method enables multiple-attributes manipulation and achieves convincing sample quality and attribute alignments, while significantly reducing computational requirements compared to pixel-based approaches for representational decoupling. Code will be released soon.

Jiang Zhou, Xiaohu Zhao, Xinwei Wu et al.

Cross-cultural entity translation remains challenging for large language models (LLMs) as literal or phonetic renderings are usually yielded instead of culturally appropriate translations in context. However, relevant knowledge may already be encoded in model parameters during large-scale pre-training. To incentivize the effective use of parametric knowledge, we propose EA-RLVR (Entity-Anchored Reinforcement Learning with Verifiable Rewards), a training framework that optimizes cross-cultural entity translation without relying on external knowledge bases. EA-RLVR anchors supervision on a verifiable, entity-level reward signal and incorporates lightweight structural gates to stabilize optimization. This design steers the model toward learning a robust reasoning process rather than merely imitating reference translations. We evaluate EA-RLVR on XC-Translate and observe consistent improvements in both entity translation accuracy and out-of-domain generalization. Specifically, training on merely 7k samples boosts Qwen3-14B's entity translation accuracy from 23.66\% to 31.87\% on a 50k test set comprising entirely unseen entities. The learned entity translation ability also transfers to general translation, yielding +1.35 XCOMET on WMT24++, which scales to +1.59 with extended optimization. Extensive analyses of $pass@k$ dynamics and reward formulations attribute these gains to superior sampling efficiency and a stable optimization landscape.

Kimi Team, Angang Du, Bofei Gao et al. · pku, tsinghua

Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).

Yangyang Liu, Yuhao Wang, Pingping Zhang

Multi-modal object Re-IDentification (ReID) is devoted to retrieving specific objects through the exploitation of complementary multi-modal image information. Existing methods mainly concentrate on the fusion of multi-modal features, yet neglecting the background interference. Besides, current multi-modal fusion methods often focus on aligning modality pairs but suffer from multi-modal consistency alignment. To address these issues, we propose a novel selective interaction and global-local alignment framework called Signal for multi-modal object ReID. Specifically, we first propose a Selective Interaction Module (SIM) to select important patch tokens with intra-modal and inter-modal information. These important patch tokens engage in the interaction with class tokens, thereby yielding more discriminative features. Then, we propose a Global Alignment Module (GAM) to simultaneously align multi-modal features by minimizing the volume of 3D polyhedra in the gramian space. Meanwhile, we propose a Local Alignment Module (LAM) to align local features in a shift-aware manner. With these modules, our proposed framework could extract more discriminative features for object ReID. Extensive experiments on three multi-modal object ReID benchmarks (i.e., RGBNT201, RGBNT100, MSVR310) validate the effectiveness of our method. The source code is available at https://github.com/010129/Signal.

Lingfeng Ming, Bo Zeng, Chenyang Lyu et al.

Large Language Models (LLMs) have achieved remarkable progress in recent years; however, their excellent performance is still largely limited to major world languages, primarily English. Many LLMs continue to face challenges with multilingual tasks, especially when it comes to low-resource languages. To address this issue, we introduced Marco-LLM: Massive multilingual training for cross-lingual enhancement LLM. We have collected a substantial amount of multilingual data for several low-resource languages and conducted extensive continual pre-training using the Qwen2 models. This effort has resulted in a multilingual LLM named Marco-LLM. Through comprehensive evaluations on various multilingual benchmarks, including MMMLU, AGIEval, Belebele, Flores-200, XCOPA and many others, Marco-LLM has demonstrated substantial improvements over state-of-the-art LLMs. Furthermore, Marco-LLM achieved substantial enhancements in any-to-any machine translation tasks, showing the effectiveness of our multilingual LLM. Marco-LLM is a pioneering multilingual LLM designed to not only perform exceptionally well in multilingual tasks, including low-resource languages, but also maintain strong performance in English and other major languages, closing the performance gap between high- and low-resource language capabilities. By bridging languages, this effort demonstrates our dedication to ensuring LLMs work accurately across various languages.

Yangyang Liu, Shoubin Li

In the rapidly evolving field of scientific research, efficiently extracting key information from the burgeoning volume of scientific papers remains a formidable challenge. This paper introduces an innovative framework designed to automate the extraction of vital data from scientific PDF documents, enabling researchers to discern future research trajectories more readily. AutoIE uniquely integrates four novel components: (1) A multi-semantic feature fusion-based approach for PDF document layout analysis; (2) Advanced functional block recognition in scientific texts; (3) A synergistic technique for extracting and correlating information on molecular sieve synthesis; (4) An online learning paradigm tailored for molecular sieve literature. Our SBERT model achieves high Marco F1 scores of 87.19 and 89.65 on CoNLL04 and ADE datasets. In addition, a practical application of AutoIE in the petrochemical molecular sieve synthesis domain demonstrates its efficacy, evidenced by an impressive 78\% accuracy rate. This research paves the way for enhanced data management and interpretation in molecular sieve synthesis. It is a valuable asset for seasoned experts and newcomers in this specialized field.

Yinwang Ren, Yangyang Liu, Tang Ji et al.

AI agents are autonomous systems designed to perceive, reason, and act within dynamic environments. With the rapid advancements in generative AI (GenAI), large language models (LLMs) and multimodal large language models (MLLMs) have significantly improved AI agents' capabilities in semantic comprehension, complex reasoning, and autonomous decision-making. At the same time, the rise of Agentic AI highlights adaptability and goal-directed autonomy in dynamic and complex environments. LLMs-based AI Agents (LLM-Agents), MLLMs-based AI Agents (MLLM-Agents), and Agentic AI contribute to expanding AI's capabilities in information processing, environmental perception, and autonomous decision-making, opening new avenues for smart manufacturing. However, the definitions, capability boundaries, and practical applications of these emerging AI paradigms in smart manufacturing remain unclear. To address this gap, this study systematically reviews the evolution of AI and AI agent technologies, examines the core concepts and technological advancements of LLM-Agents, MLLM-Agents, and Agentic AI, and explores their potential applications in and integration into manufacturing, along with the potential challenges they may face.

Shibo Gao, Peipei Yang, Haiyang Guo et al.

Video anomaly detection (VAD) aims to identify and ground anomalous behaviors or events in videos, serving as a core technology in the fields of intelligent surveillance and public safety. With the advancement of deep learning, the continuous evolution of deep model architectures has driven innovation in VAD methodologies, significantly enhancing feature representation and scene adaptability, thereby improving algorithm generalization and expanding application boundaries. More importantly, the rapid development of multi-modal large language (MLLMs) and large language models (LLMs) has introduced new opportunities and challenges to the VAD field. Under the support of MLLMs and LLMs, VAD has undergone significant transformations in terms of data annotation, input modalities, model architectures, and task objectives. The surge in publications and the evolution of tasks have created an urgent need for systematic reviews of recent advancements. This paper presents the first comprehensive survey analyzing VAD methods based on MLLMs and LLMs, providing an in-depth discussion of the changes occurring in the VAD field in the era of large models and their underlying causes. Additionally, this paper proposes a unified framework that encompasses both deep neural network (DNN)-based and LLM-based VAD methods, offering a thorough analysis of the new VAD paradigms empowered by LLMs, constructing a classification system, and comparing their strengths and weaknesses. Building on this foundation, this paper focuses on current VAD methods based on MLLMs/LLMs. Finally, based on the trajectory of technological advancements and existing bottlenecks, this paper distills key challenges and outlines future research directions, offering guidance for the VAD community.

Shibo Gao, Peipei Yang, Yangyang Liu et al.

Video Anomaly Detection (VAD) aims to identify anomalous events in videos and accurately determine their time intervals. Current VAD methods mainly fall into two categories: traditional DNN-based approaches that focus on temporal localization, and LLM-based approaches that emphasize semantic understanding. Both anomaly understanding and grounding are essential for comprehensive video anomaly detection and can complement each other. However, no existing model or dataset supports both tasks simultaneously. To address this, we introduce VAGU (Video Anomaly Grounding and Understanding), the first benchmark to integrate both tasks. Each VAGU instance includes annotations for anomaly category, semantic explanation, precise temporal grounding and Video QA. We also provide multiple-choice Video QA for objective evaluation. Based on this dataset, we propose Glance then Scrutinize (GtS), a training-free framework guided by textual prompts. The framework first enables coarse localization of high-probability anomalous regions, followed by detailed anomaly interpretation and temporal boundary refinement. Additionally, we propose the JeAUG metric, which jointly evaluates semantic interpretability and temporal precision, overcoming the limitations of traditional metrics. Extensive experiments verify the effectiveness of our benchmark, framework, and evaluation metric.

Hao Wang, Linlong Xu, Heng Liu et al.

Direct Preference Optimization (DPO) is a powerful paradigm for aligning Large Language Models (LLMs) to human preferences in Machine Translation (MT), but current methods are hindered by two fundamental challenges: (1) flawed reward signals from Quality Estimation (QE) models that overlook critical errors like translation hallucination, and (2) inefficient data utilization that discards valuable learning signals by selecting only a single win-loss pair. To address these limitations, we introduce M^2PO: Multi-Pair, Multi-Perspective Preference Optimization. Our framework integrates a multi-perspective reward engine that creates a more robust signal by combining two key viewpoints: a new hallucination penalty for factuality, and an innovative dynamic quality score that adaptively fuses external evaluations with the model's own evolving judgment. This is synergistically paired with a multi-pair construction strategy that systematically creates a comprehensive set of preference pairs from the entire pool of translation candidates. This synergistic approach ensures the model learns from a richer spectrum of quality trade-offs, leading to more robust and faithful translations. On challenging WMT21-22 benchmarks, M^2PO substantially outperforms existing preference optimization methods and demonstrates highly competitive performance against leading proprietary LLMs.

Zhiqing Wei, Fengkai Zhang, Shuo Chang et al.

With autonomous driving developing in a booming stage, accurate object detection in complex scenarios attract wide attention to ensure the safety of autonomous driving. Millimeter wave (mmWave) radar and vision fusion is a mainstream solution for accurate obstacle detection. This article presents a detailed survey on mmWave radar and vision fusion based obstacle detection methods. First, we introduce the tasks, evaluation criteria, and datasets of object detection for autonomous driving. The process of mmWave radar and vision fusion is then divided into three parts: sensor deployment, sensor calibration, and sensor fusion, which are reviewed comprehensively. Specifically, we classify the fusion methods into data level, decision level, and feature level fusion methods. In addition, we introduce three-dimensional(3D) object detection, the fusion of lidar and vision in autonomous driving and multimodal information fusion, which are promising for the future. Finally, we summarize this article.

Guangxia Lia, Yulong Shena, Peilin Zhaob et al.

Industrial control systems are critical to the operation of industrial facilities, especially for critical infrastructures, such as refineries, power grids, and transportation systems. Similar to other information systems, a significant threat to industrial control systems is the attack from cyberspace---the offensive maneuvers launched by "anonymous" in the digital world that target computer-based assets with the goal of compromising a system's functions or probing for information. Owing to the importance of industrial control systems, and the possibly devastating consequences of being attacked, significant endeavors have been attempted to secure industrial control systems from cyberattacks. Among them are intrusion detection systems that serve as the first line of defense by monitoring and reporting potentially malicious activities. Classical machine-learning-based intrusion detection methods usually generate prediction models by learning modest-sized training samples all at once. Such approach is not always applicable to industrial control systems, as industrial control systems must process continuous control commands with limited computational resources in a nonstop way. To satisfy such requirements, we propose using online learning to learn prediction models from the controlling data stream. We introduce several state-of-the-art online learning algorithms categorically, and illustrate their efficacies on two typically used testbeds---power system and gas pipeline. Further, we explore a new cost-sensitive online learning algorithm to solve the class-imbalance problem that is pervasive in industrial intrusion detection systems. Our experimental results indicate that the proposed algorithm can achieve an overall improvement in the detection rate of cyberattacks in industrial control systems.