Xiaodong Cai

Meituan LongCat Team, Anchun Gui, Bei Li et al.

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

Yuxin Chen, Xiaodong Cai, Junfeng Fang et al.

Recent advances in large language models (LLMs) have facilitated the widespread deployment of LLMs as interactive agents capable of reasoning, planning, and tool use. Despite strong performance on existing benchmarks, such agents often exhibit notable degradation when deployed in real-world settings, where environments are inherently stochastic and imperfect. We argue that this discrepancy arises from a fundamental mismatch between idealized training settings and real-world interaction dynamics, where current paradigms rely on carefully curated task instructions and stable, well-controlled environments. To address this gap, we propose NoisyAgent, an agentic training framework that explicitly incorporates environmental imperfections into the agent learning process. We identify two major sources of interaction noise in real-world scenarios: user noise, which captures ambiguity and variability in user interaction, and tool noise, which reflects failures and anomalies in tool execution. We introduce such perturbations into the training pipeline by modifying user interaction patterns and simulating tool execution results within the training environment. To stabilize training while encouraging agents to handle increasingly challenging imperfections, noise is applied to only a subset of rollouts and progressively increased in difficulty as the model adapts to the current noise level. Extensive experiments demonstrate that our approach consistently improves agent robustness under noisy and dynamic environments. Our analysis reveals that training under noise conditions also yields performance gains on idealized benchmarks, suggesting that controlled exposure to environmental noise promotes more generalizable reasoning and decision-making behaviors. Our findings highlight the importance of modeling interaction imperfections for bridging the gap between agent training and real-world deployment.

Anchen Sun, Elizabeth J. Franzmann, Zhibin Chen et al.

Recent advancements in image classification have demonstrated that contrastive learning (CL) can aid in further learning tasks by acquiring good feature representation from a limited number of data samples. In this paper, we applied CL to tumor transcriptomes and clinical data to learn feature representations in a low-dimensional space. We then utilized these learned features to train a classifier to categorize tumors into a high- or low-risk group of recurrence. Using data from The Cancer Genome Atlas (TCGA), we demonstrated that CL can significantly improve classification accuracy. Specifically, our CL-based classifiers achieved an area under the receiver operating characteristic curve (AUC) greater than 0.8 for 14 types of cancer, and an AUC greater than 0.9 for 2 types of cancer. We also developed CL-based Cox (CLCox) models for predicting cancer prognosis. Our CLCox models trained with the TCGA data outperformed existing methods significantly in predicting the prognosis of 19 types of cancer under consideration. The performance of CLCox models and CL-based classifiers trained with TCGA lung and prostate cancer data were validated using the data from two independent cohorts. We also show that the CLCox model trained with the whole transcriptome significantly outperforms the Cox model trained with the 21 genes of Oncotype DX that is in clinical use for breast cancer patients. CL-based classifiers and CLCox models for 19 types of cancer are publicly available and can be used to predict cancer prognosis using the RNA-seq transcriptome of an individual tumor. Python codes for model training and testing are also publicly accessible, and can be applied to train new CL-based models using gene expression data of tumors.

Xiaodong Cai, Hai Lin, Shaoxiong Zhan et al.

Token sampling strategies critically influence text generation quality in large language models (LLMs). However, existing methods introduce additional hyperparameters, requiring extensive tuning and complicating deployment. We present Entropy Equilibrium Sampling (EES), an auxiliary hyperparameter-free approach inspired by information theory that can dynamically adjust candidate sets by balancing normalized entropy with probability mass. We evaluate EES on both reasoning and generation tasks across a range of model architectures. Our results show that EES consistently performs well across temperature settings, delivering competitive accuracy and coherence while maintaining diversity. By eliminating the need for hyperparameter tuning, EES greatly simplifies deployment while improving performance. Code is available at https://github.com/shuanncai/EES

Meituan LongCat Team, Bayan, Bei Li et al.

We introduce LongCat-Flash, a 560-billion-parameter Mixture-of-Experts (MoE) language model designed for both computational efficiency and advanced agentic capabilities. Stemming from the need for scalable efficiency, LongCat-Flash adopts two novel designs: (a) Zero-computation Experts, which enables dynamic computational budget allocation and activates 18.6B-31.3B (27B on average) per token depending on contextual demands, optimizing resource usage. (b) Shortcut-connected MoE, which enlarges the computation-communication overlap window, demonstrating notable gains in inference efficiency and throughput compared to models of a comparable scale. We develop a comprehensive scaling framework for large models that combines hyperparameter transfer, model-growth initialization, a multi-pronged stability suite, and deterministic computation to achieve stable and reproducible training. Notably, leveraging the synergy among scalable architectural design and infrastructure efforts, we complete model training on more than 20 trillion tokens within 30 days, while achieving over 100 tokens per second (TPS) for inference at a cost of \$0.70 per million output tokens. To cultivate LongCat-Flash towards agentic intelligence, we conduct a large-scale pre-training on optimized mixtures, followed by targeted mid- and post-training on reasoning, code, and instructions, with further augmentation from synthetic data and tool use tasks. Comprehensive evaluations demonstrate that, as a non-thinking foundation model, LongCat-Flash delivers highly competitive performance among other leading models, with exceptional strengths in agentic tasks. The model checkpoint of LongCat-Flash is open-sourced to foster community research. LongCat Chat: https://longcat.ai Hugging Face: https://huggingface.co/meituan-longcat GitHub: https://github.com/meituan-longcat

Ziyuan Xia, Anchen Sun, Xiaodong Cai et al.

Environmental, social, and governance (ESG) reports are globally recognized as a keystone in sustainable enterprise development. However, current literature has not concluded the development of topics and trends in ESG contexts in the twenty-first century. Therefore, We selected 1114 ESG reports from firms in the technology industry to analyze the evolutionary trends of ESG topics by text mining. We discovered the homogenization effect towards low environmental, medium governance, and high social features in the evolution. We also designed a strategic framework to look closer into the dynamic changes of firms' within-industry scores and across-domain importances. We found that companies are gradually converging towards the third quadrant, which indicates that firms contribute less to industrial outstanding and professional distinctiveness in ESG reporting. Firms choose to imitate ESG reports from each other to mitigate uncertainty and enhance behavioral legitimacy.

Anchen Sun, Zhibin Chen, Xiaodong Cai

The Cox Proportional Hazards (PH) model is widely used in survival analysis. Recently, artificial neural network (ANN)-based Cox-PH models have been developed. However, training these Cox models with high-dimensional features typically requires a substantial number of labeled samples containing information about time-to-event. The limited availability of labeled data for training often constrains the performance of ANN-based Cox models. To address this issue, we employed a deep semi-supervised learning (DSSL) approach to develop single- and multi-modal ANN-based Cox models based on the Mean Teacher (MT) framework, which utilizes both labeled and unlabeled data for training. We applied our model, named Cox-MT, to predict the prognosis of several types of cancer using data from The Cancer Genome Atlas (TCGA). Our single-modal Cox-MT models, utilizing TCGA RNA-seq data or whole slide images, significantly outperformed the existing ANN-based Cox model, Cox-nnet, using the same data set across four types of cancer considered. As the number of unlabeled samples increased, the performance of Cox-MT significantly improved with a given set of labeled data. Furthermore, our multi-modal Cox-MT model demonstrated considerably better performance than the single-modal model. In summary, the Cox-MT model effectively leverages both labeled and unlabeled data to significantly enhance prediction accuracy compared to existing ANN-based Cox models trained solely on labeled data.

Ruipeng Wang, Yuxin Chen, Yukai Wang et al.

Recent advances in large language models have enabled LLM-based agents to achieve strong performance on a variety of benchmarks. However, their performance in real-world deployments often that observed on benchmark settings, especially in complex and imperfect environments. This discrepancy largely arises because prevailing training and evaluation paradigms are typically built on idealized assumptions, overlooking the inherent stochasticity and noise present in real-world interactions. To bridge this gap, we introduce AgentNoiseBench, a framework for systematically evaluating the robustness of agentic models under noisy environments. We first conduct an in-depth analysis of biases and uncertainties in real-world scenarios and categorize environmental noise into two primary types: user-noise and tool-noise. Building on this analysis, we develop an automated pipeline that injects controllable noise into existing agent-centric benchmarks while preserving task solvability. Leveraging this pipeline, we perform extensive evaluations across a wide range of models with diverse architectures and parameter scales. Our results reveal consistent performance variations under different noise conditions, highlighting the sensitivity of current agentic models to realistic environmental perturbations.

Shaoxiong Zhan, Yanlin Lai, Zheng Liu et al.

Current Large Language Models have achieved Olympiad-level logic, yet Vision-Language Models paradoxically falter on elementary spatial tasks like block counting. This capability mismatch reveals a critical ``spatial intelligence gap,'' where models fail to construct coherent 3D mental representations from 2D observations. We uncover this gap via diagnostic analyses showing the bottleneck is a missing view-consistent spatial interface rather than insufficient visual features or weak reasoning. To bridge this, we introduce \textbf{3ViewSense}, a framework that grounds spatial reasoning in Orthographic Views. Drawing on engineering cognition, we propose a ``Simulate-and-Reason'' mechanism that decomposes complex scenes into canonical orthographic projections to resolve geometric ambiguities. By aligning egocentric perceptions with these allocentric references, our method facilitates explicit mental rotation and reconstruction. Empirical results on spatial reasoning benchmarks demonstrate that our method significantly outperforms existing baselines, with consistent gains on occlusion-heavy counting and view-consistent spatial reasoning. The framework also improves the stability and consistency of spatial descriptions, offering a scalable path toward stronger spatial intelligence in multimodal systems.

Wei He, Yueqing Sun, Hongyan Hao et al.

As LLM-based agents are increasingly deployed in real-life scenarios, existing benchmarks fail to capture their inherent complexity of handling extensive information, leveraging diverse resources, and managing dynamic user interactions. To address this gap, we introduce VitaBench, a challenging benchmark that evaluates agents on versatile interactive tasks grounded in real-world settings. Drawing from daily applications in food delivery, in-store consumption, and online travel services, VitaBench presents agents with the most complex life-serving simulation environment to date, comprising 66 tools. Through a framework that eliminates domain-specific policies, we enable flexible composition of these scenarios and tools, yielding 100 cross-scenario tasks (main results) and 300 single-scenario tasks. Each task is derived from multiple real user requests and requires agents to reason across temporal and spatial dimensions, utilize complex tool sets, proactively clarify ambiguous instructions, and track shifting user intent throughout multi-turn conversations. Moreover, we propose a rubric-based sliding window evaluator, enabling robust assessment of diverse solution pathways in complex environments and stochastic interactions. Our comprehensive evaluation reveals that even the most advanced models achieve only 30% success rate on cross-scenario tasks, and less than 50% success rate on others. Overall, we believe VitaBench will serve as a valuable resource for advancing the development of AI agents in practical real-world applications. The code, dataset, and leaderboard are available at https://vitabench.github.io/

Xin Zhou, Chunlei Du, Xiaodong Cai

Cluster analysis organizes data into sensible groupings and is one of fundamental modes of understanding and learning. The widely used K-means and hierarchical clustering methods can be dramatically suboptimal due to local minima. Recently introduced convex clustering approach formulates clustering as a convex optimization problem and ensures a globally optimal solution. However, the state-of-the-art convex clustering algorithms, based on the alternating direction method of multipliers (ADMM) or the alternating minimization algorithm (AMA), require large computation and memory space, which limits their applications. In this paper, we develop a very efficient smoothing proximal gradient algorithm (Sproga) for convex clustering. Our Sproga is faster than ADMM- or AMA-based convex clustering algorithms by one to two orders of magnitude. The memory space required by Sproga is less than that required by ADMM and AMA by at least one order of magnitude. Computer simulations and real data analysis show that Sproga outperforms several well known clustering algorithms including K-means and hierarchical clustering. The efficiency and superior performance of our algorithm will help convex clustering to find its wide application.

Xiaodong Cai, Jingyi Ma, Wei Liu et al.

FMCW radar could detect object's range, speed and Angleof-Arrival, advantages are robust to bad weather, good range resolution, and good speed resolution. In this paper, we consider the FMCW radar as a novel interacting interface on laptop. We merge sequences of object's range, speed, azimuth information into single input, then feed to a convolution neural network to learn spatial and temporal patterns. Our model achieved 96% accuracy on test set and real-time test.