虽然人类在获得语言之前很久就发展出了核心视觉技能，但当代多模态语言模型（MLLMs）仍然严重依赖语言先验来弥补其脆弱的视觉理解能力。我们发现了一个关键事实：最先进的MLLMs在人类（甚至3岁儿童）能轻松解决的基本视觉任务上持续失败。为系统研究这一差距，我们引入了BabyVision，一个旨在评估MLLMs独立于语言知识的核心视觉能力的基准。BabyVision涵盖广泛任务，有388个项目分为四个关键类别的22个子类。实证结果和人工评估表明，领先的MLLMs表现显著低于人类基线。Gemini3 - Pro - Preview得分为49.7，落后于6岁儿童，远低于成人平均得分94.1。这些结果表明，尽管在知识密集型评估中表现出色，但当前MLLMs仍缺乏基本视觉原语。BabyVision的进展代表了向人类水平视觉感知和推理能力迈出的一步。我们还通过提出BabyVision - Gen和自动评估工具包探索用生成模型解决视觉推理问题。我们的代码和基准数据在该https网址发布以供复制。
英文摘要
While humans develop core visual skills long before acquiring language, contemporary Multimodal LLMs (MLLMs) still rely heavily on linguistic priors to compensate for their fragile visual understanding. We uncovered a crucial fact: state-of-the-art MLLMs consistently fail on basic visual tasks that humans, even 3-year-olds, can solve effortlessly. To systematically investigate this gap, we introduce BabyVision, a benchmark designed to assess core visual abilities independent of linguistic knowledge for MLLMs. BabyVision spans a wide range of tasks, with 388 items divided into 22 subclasses across four key categories. Empirical results and human evaluation reveal that leading MLLMs perform significantly below human baselines. Gemini3-Pro-Preview scores 49.7, lagging behind 6-year-old humans and falling well behind the average adult score of 94.1. These results show despite excelling in knowledge-heavy evaluations, current MLLMs still lack fundamental visual primitives. Progress in BabyVision represents a step toward human-level visual perception and reasoning capabilities. We also explore solving visual reasoning with generation models by proposing BabyVision-Gen and automatic evaluation toolkit. Our code and benchmark data are released at https://github.com/UniPat-AI/BabyVision for reproduction.
In large-scale optimization, the cheapness and effectiveness of update steps are the most crucial factors for a successful optimizer. Sign-based optimizers like Lion or Signum produce cheap per-step updates, whereas Muon's spectral matrix-sign update gives a much stronger direction at a substantially higher per-step cost. In this work, we propose LionMuon, which retains the effectiveness of Muon steps while considerably cutting the averaged iteration cost, similar to sign-based methods. It alternates between Lion's and Muon's updates on a fixed period P, sharing a single dual-EMA momentum buffer between them. The optimizer state memory therefore matches Lion and is exactly half of AdamW's. A simpler single-EMA variant, SignMuon, by itself already outperforms pure Muon. At P = 2, LionMuon Pareto-dominates Muon, Lion, Signum, and AdamW on every dataset and architecture we tested at 124M model size, reaching lower validation loss at lower compute, and the same advantage persists at 355M and 720M scale. On the theory side, we prove sharp complexity bounds under heavy-tailed noise which are governed by period-averaged smoothness and noise that interpolate between Muon's and Lion's constants. These bounds predict the compute-optimal period and the conditions under which LionMuon outruns Muon and Lion. Code: https://github.com/brain-lab-research/lion-muon
Optimization problems are central to decision-making in manufacturing, logistics, scheduling, and other industrial settings. Translating complicated descriptions of these problems into solver-ready formulations requires specialized operations research (OR) expertise, making it hard to scale. We present AutoOR, a scalable synthetic data generation and reinforcement learning pipeline that trains LLMs to autoformalize optimization problems specified in natural language across linear, mixed-integer, and non-linear categories. AutoOR generates verified training data from standard optimization forms and uses solver execution feedback as the reward signal for RL post-training. AutoOR applied to an 8B model achieves state-of-the-art or competitive results across six established OR benchmarks, matching significantly larger frontier models. For a non-linear problem class involving physical dynamics, where frontier models score near 0%, we introduce a curriculum RL strategy that bootstraps from limited initial training data to make this class tractable for post-training. We believe that methods such as AutoOR can significantly accelerate industrial decision-making with AI.
OPRIDE: Offline Preference-based Reinforcement Learning via In-Dataset Exploration
OPRIDE: 通过数据集内探索实现离线偏好驱动强化学习
Yiqin Yang, Hao Hu, Yihuan Mao, Jin Zhang, Chengjie Wu, Yuhua Jiang, Xu Yang, Runpeng Xie, Yi Fan, Bo Liu, Yang Gao, Bo Xu, Chongjie Zhang
机构
*
The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所复杂系统认知与决策智能重点实验室）
；
Tsinghua University（清华大学）
；
Moonshot AI（月之暗面）
；
Amazon（亚马逊）
；
University of Arizona（亚利桑那大学）
；
Washington University in St. Louis（圣路易斯华盛顿大学）
Preference-based reinforcement learning (PbRL) can help avoid sophisticated reward designs and align better with human intentions, showing great promise in various real-world applications. However, obtaining human feedback for preferences can be expensive and time-consuming, which forms a strong barrier for PbRL. In this work, we address the problem of low query efficiency in offline PbRL, pinpointing two primary reasons: inefficient exploration and overoptimization of learned reward functions. In response to these challenges, we propose a novel algorithm, \textbf{O}ffline \textbf{P}b\textbf{R}L via \textbf{I}n-\textbf{D}ataset \textbf{E}xploration (OPRIDE), designed to enhance the query efficiency of offline PbRL. OPRIDE consists of two key features: a principled exploration strategy that maximizes the informativeness of the queries and a discount scheduling mechanism aimed at mitigating overoptimization of the learned reward functions. Through empirical evaluations, we demonstrate that OPRIDE significantly outperforms prior methods, achieving strong performance with notably fewer queries. Moreover, we provide theoretical guarantees of the algorithm's efficiency. Experimental results across various locomotion, manipulation, and navigation tasks underscore the efficacy and versatility of our approach.
Reinforcement Learning (RL) has emerged as a critical technique for advancing modern Large Language Models (LLMs), yet existing synchronous RL systems face severe performance bottlenecks. The rollout phase, which dominates end-to-end iteration time, suffers from substantial long-tail latency and poor resource utilization due to inherent workload imbalance. We present Seer, a novel context learning RL system that addresses these challenges through a key observation: requests sharing the same prompt exhibit strong similarities in output lengths and response patterns. Leveraging this insight, Seer introduces three coordinated techniques: (1) divided rollout for dynamic load balancing, (2) context-aware scheduling to mitigate long-tail request delays, and (3) adaptive grouped speculative decoding to accelerate generation. These mechanisms work in concert to markedly reduce long-tail latency and improve resource efficiency during rollout. Evaluations on production-grade RL workloads demonstrate that Seer achieves up to 2.04$\times$ end-to-end rollout throughput improvement compared to the state-of-the-art synchronous RL systems, while notably reducing long-tail latency by 72-94%.
VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning?
VideoReasonBench: 大规模语言模型能否进行以视觉为中心的复杂视频推理？
Yuanxin Liu, Kun Ouyang, Haoning Wu, Yi Liu, Lin Sui, Xinhao Li, Yan Zhong, Y. Charles, Xinyu Zhou, Xu Sun
机构
*
State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University（多媒体信息处理国家重点实验室，计算机科学学院，北京大学）
；
Moonshot AI
；
Nanjing University（南京大学）
；
School of Mathematical Sciences, Peking University（数学科学学院，北京大学）
Recent studies have shown that long chain-of-thought (CoT) reasoning can significantly enhance the performance of large language models (LLMs) on complex tasks. However, this benefit is yet to be demonstrated in the domain of video understanding, since most existing benchmarks lack the reasoning depth required to demonstrate the advantages of extended CoT chains. While recent efforts have proposed benchmarks aimed at video reasoning, the tasks are often knowledge-driven and do not rely heavily on visual content. To bridge this gap, we introduce VideoReasonBench, a benchmark designed to evaluate vision-centric, complex video reasoning. To ensure visual richness and high reasoning complexity, each video in VideoReasonBench depicts a sequence of fine-grained operations on a latent state that is only visible in part of the video. The questions evaluate three escalating levels of video reasoning skills: recalling observed visual information, inferring the content of latent states, and predicting information beyond the video. Under such task setting, models have to precisely recall multiple operations in the video, and perform step-by-step reasoning to get correct final answers for these questions. Using VideoReasonBench, we comprehensively evaluate 18 state-of-the-art multimodal LLMs (MLLMs), finding that most perform poorly on complex video reasoning -- e.g., GPT-4o achieves only 6.9% accuracy -- while the thinking-enhanced Gemini-2.5-Pro significantly outperforms others with 56.0% accuracy. Our investigations on "test-time scaling" further reveal that extended thinking budget, while offering none or minimal benefits on existing video benchmarks, is essential for improving the performance on VideoReasonBench.
机构
*
Beijing University of Posts and Telecommunications（北京邮电大学）
；
Tsinghua University（清华大学）
；
University of Chinese Academy of Sciences（中国科学院大学）
；
Peking University（北京大学）
；
Moonshot AI
Recent advancements in large language models (LLMs) have demonstrated significant progress in math and code reasoning capabilities. However, existing code benchmark are limited in their ability to evaluate the full spectrum of these capabilities, particularly at the competitive level. To bridge this gap, we introduce OJBench, a novel and challenging benchmark designed to assess the competitive-level code reasoning abilities of LLMs. OJBench comprises 232 programming competition problems from NOI and ICPC, providing a more rigorous test of models' reasoning skills. We conducted a comprehensive evaluation using OJBench on 37 models, including both closed-source and open-source models, reasoning-oriented and non-reasoning-oriented models. Our results indicate that even state-of-the-art reasoning-oriented models, such as o4-mini and Gemini-2.5-pro-exp, struggle with highly challenging competition-level problems. This highlights the significant challenges that models face in competitive-level code reasoning.
Affinity Contrastive Learning for Skeleton-based Human Activity Understanding
基于骨架的人体活动理解的亲和对比学习
Hongda Liu, Yunfan Liu, Min Ren, Lin Sui, Yunlong Wang, Zhenan Sun
机构
*
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）
；
School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）
；
School of Electronic, Electrical and Communication Engineering, Univeristy of Chinese Academy of Sciences（中国科学院大学电子电气与通信工程学院）
；
School of Artificial Intelligence, Beijing University of Posts and Telecommunications（北京邮电大学人工智能学院）
；
Moonshot AI
AI总结
ACLNet通过引入亲和度度量和动态温度调度，提升基于骨架的人体活动理解的特征判别能力。
CommentsAccepted by TBIOM
Journal refIEEE Transactions on Biometrics, Behavior, and Identity Science (2026)
In skeleton-based human activity understanding, existing methods often adopt the contrastive learning paradigm to construct a discriminative feature space. However, many of these approaches fail to exploit the structural inter-class similarities and overlook the impact of anomalous positive samples. In this study, we introduce ACLNet, an Affinity Contrastive Learning Network that explores the intricate clustering relationships among human activity classes to improve feature discrimination. Specifically, we propose an affinity metric to refine similarity measurements, thereby forming activity superclasses that provide more informative contrastive signals. A dynamic temperature schedule is also introduced to adaptively adjust the penalty strength for various superclasses. In addition, we employ a margin-based contrastive strategy to improve the separation of hard positive and negative samples within classes. Extensive experiments on NTU RGB+D 60, NTU RGB+D 120, Kinetics-Skeleton, PKU-MMD, FineGYM, and CASIA-B demonstrate the superiority of our method in skeleton-based action recognition, gait recognition, and person re-identification. The source code is available at https://github.com/firework8/ACLNet.
机构
*
The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences（认知与决策智能复杂系统重点实验室，自动化研究所，中国科学院）
；
Tsinghua University（清华大学）
；
Moonshot AI
；
Nankai University（南开大学）
；
Faculty of Computing, Harbin Institute of Technology（哈尔滨工业大学计算机学院）
In the realm of multi-agent systems, the challenge of \emph{partial observability} is a critical barrier to effective coordination and decision-making. Existing approaches, such as belief state estimation and inter-agent communication, often fall short. Belief-based methods are limited by their focus on past experiences without fully leveraging global information, while communication methods often lack a robust model to effectively utilize the auxiliary information they provide. To solve this issue, we propose Global State Diffusion Algorithm~(GlobeDiff) to infer the global state based on the local observations. By formulating the state inference process as a multi-modal diffusion process, GlobeDiff overcomes ambiguities in state estimation while simultaneously inferring the global state with high fidelity. We prove that the estimation error of GlobeDiff under both unimodal and multi-modal distributions can be bounded. Extensive experimental results demonstrate that GlobeDiff achieves superior performance and is capable of accurately inferring the global state.
We introduce WorldVQA, a benchmark designed to evaluate the atomic visual world knowledge of Multimodal Large Language Models (MLLMs). Unlike current evaluations, which often conflate visual knowledge retrieval with reasoning, WorldVQA decouples these capabilities to strictly measure "what the model memorizes." The benchmark assesses the atomic capability of grounding and naming visual entities across a stratified taxonomy, spanning from common head-class objects to long-tail rarities. We expect WorldVQA to serve as a rigorous test for visual factuality, thereby establishing a standard for assessing the encyclopedic breadth and hallucination rates of current and next-generation frontier models.
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.
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.
Large Language Models (LLMs) are increasingly applied to software engineering (SWE), with SWE-bench as a key benchmark. Solutions are split into SWE-Agent frameworks with multi-turn interactions and workflow-based Agentless methods with single-turn verifiable steps. We argue these paradigms are not mutually exclusive: reasoning-intensive Agentless training induces skill priors, including localization, code edit, and self-reflection that enable efficient and effective SWE-Agent adaptation. In this work, we first curate the Agentless training recipe and present Kimi-Dev, an open-source SWE LLM achieving 60.4\% on SWE-bench Verified, the best among workflow approaches. With additional SFT adaptation on 5k publicly-available trajectories, Kimi-Dev powers SWE-Agents to 48.6\% pass@1, on par with that of Claude 3.5 Sonnet (241022 version). These results show that structured skill priors from Agentless training can bridge workflow and agentic frameworks for transferable coding agents.
DAIL: Beyond Task Ambiguity for Language-Conditioned Reinforcement Learning
Runpeng Xie, Quanwei Wang, Hao Hu, Zherui Zhou, Ni Mu, Xiyun Li, Yiqin Yang, Shuang Xu, Qianchuan Zhao, Bo XU
机构
*
The Key Laboratory of Cognition and Decision Intelligence for Complex Systems Institute of Automation Chinese Academy of Sciences（认知与决策智能复杂系统重点实验室，自动化研究所，中国科学院）
；
Department of Automation Tsinghua University（清华大学自动化系）
；
Moonshot AI
；
Department of Computer Science and Engineering Washington University（华盛顿大学计算机科学与工程系）
；
Tecent AI Lab（腾讯AI实验室）
Comprehending natural language and following human instructions are critical capabilities for intelligent agents. However, the flexibility of linguistic instructions induces substantial ambiguity across language-conditioned tasks, severely degrading algorithmic performance. To address these limitations, we present a novel method named DAIL (Distributional Aligned Learning), featuring two key components: distributional policy and semantic alignment. Specifically, we provide theoretical results that the value distribution estimation mechanism enhances task differentiability. Meanwhile, the semantic alignment module captures the correspondence between trajectories and linguistic instructions. Extensive experimental results on both structured and visual observation benchmarks demonstrate that DAIL effectively resolves instruction ambiguities, achieving superior performance to baseline methods. Our implementation is available at https://github.com/RunpengXie/Distributional-Aligned-Learning.
机构
*
XLANG Lab, The University of Hong Kong（香港大学XLANG实验室）
；
Moonshot AI
；
Stanford University（斯坦福大学）
；
University of Waterloo（滑铁卢大学）
；
Carnegie Mellon University（卡内基梅隆大学）
CommentsUpdata author list, modify first page format, correct typos
详情
英文摘要
Vision-language models have demonstrated impressive capabilities as computer-use agents (CUAs) capable of automating diverse computer tasks. As their commercial potential grows, critical details of the most capable CUA systems remain closed. As these agents will increasingly mediate digital interactions and execute consequential decisions on our behalf, the research community needs access to open CUA frameworks to study their capabilities, limitations, and risks. To bridge this gap, we propose OpenCUA, a comprehensive open-source framework for scaling CUA data and foundation models. Our framework consists of: (1) an annotation infrastructure that seamlessly captures human computer-use demonstrations; (2) AgentNet, the first large-scale computer-use task dataset spanning 3 operating systems and 200+ applications and websites; (3) a scalable pipeline that transforms demonstrations into state-action pairs with reflective long Chain-of-Thought reasoning that sustain robust performance gains as data scales. Our end-to-end agent models demonstrate strong performance across CUA benchmarks. In particular, OpenCUA-72B achieves an average success rate of 45.0% on OSWorld-Verified, establishing a new state-of-the-art (SOTA) among open-source models. Further analysis confirms that our approach generalizes well across domains and benefits significantly from increased test-time computation. We release our annotation tool, datasets, code, and models to build open foundations for further CUA research.
Mooncake is the serving platform for Kimi, a leading LLM service provided by Moonshot AI. It features a KVCache-centric disaggregated architecture that separates the prefill and decoding clusters. It also leverages the underutilized CPU, DRAM, and SSD resources of the GPU cluster to implement a disaggregated cache of KVCache. The core of Mooncake is its KVCache-centric scheduler, which balances maximizing overall effective throughput while meeting latency-related Service Level Objectives (SLOs). Unlike traditional studies that assume all requests will be processed, Mooncake faces challenges due to highly overloaded scenarios. To mitigate these, we developed a prediction-based early rejection policy. Experiments show that Mooncake excels in long-context scenarios. Compared to the baseline method, Mooncake can achieve up to a 525% increase in throughput in certain simulated scenarios while adhering to SLOs. Under real workloads, Mooncake's innovative architecture enables Kimi to handle 75% more requests.
LeanGeo: Formalizing Competitional Geometry problems in Lean
Chendong Song, Zihan Wang, Frederick Pu, Haiming Wang, Xiaohan Lin, Junqi Liu, Jia Li, Zhengying Liu
机构
*
Moonshot AI
；
Numina
；
Peking University（北京大学）
；
University of Toronto（多伦多大学）
Comments28 pages
详情
英文摘要
Geometry problems are a crucial testbed for AI reasoning capabilities. Most existing geometry solving systems cannot express problems within a unified framework, thus are difficult to integrate with other mathematical fields. Besides, since most geometric proofs rely on intuitive diagrams, verifying geometry problems is particularly challenging. To address these gaps, we introduce LeanGeo, a unified formal system for formalizing and solving competition-level geometry problems within the Lean 4 theorem prover. LeanGeo features a comprehensive library of high-level geometric theorems with Lean's foundational logic, enabling rigorous proof verification and seamless integration with Mathlib. We also present LeanGeo-Bench, a formal geometry benchmark in LeanGeo, comprising problems from the International Mathematical Olympiad (IMO) and other advanced sources. Our evaluation demonstrates the capabilities and limitations of state-of-the-art Large Language Models on this benchmark, highlighting the need for further advancements in automated geometric reasoning. We open source the theorem library and the benchmark of LeanGeo at https://github.com/project-numina/LeanGeo/tree/master.
Scalable Geospatial Data Generation Using AlphaEarth Foundations Model
Luc Houriez, Sebastian Pilarski, Behzad Vahedi, Ali Ahmadalipour, Teo Honda Scully, Nicholas Aflitto, David Andre, Caroline Jaffe, Martha Wedner, Rich Mazzola, Josh Jeffery, Ben Messinger, Sage McGinley-Smith, Sarah Russell
High-quality labeled geospatial datasets are essential for extracting insights and understanding our planet. Unfortunately, these datasets often do not span the entire globe and are limited to certain geographic regions where data was collected. Google DeepMind's recently released AlphaEarth Foundations (AEF) provides an information-dense global geospatial representation designed to serve as a useful input across a wide gamut of tasks. In this article we propose and evaluate a methodology which leverages AEF to extend geospatial labeled datasets beyond their initial geographic regions. We show that even basic models like random forests or logistic regression can be used to accomplish this task. We investigate a case study of extending LANDFIRE's Existing Vegetation Type (EVT) dataset beyond the USA into Canada at two levels of granularity: EvtPhys (13 classes) and EvtGp (80 classes). Qualitatively, for EvtPhys, model predictions align with ground truth. Trained models achieve 81% and 73% classification accuracy on EvtPhys validation sets in the USA and Canada, despite discussed limitations.
