Vision-Language-Action (VLA) models have recently demonstrated strong performance across embodied tasks. Modern VLAs commonly employ diffusion action experts to efficiently generate high-precision continuous action chunks, while auto-regressive generation can be slower and less accurate at low-level control. Yet auto-regressive paradigms still provide complementary priors that can improve robustness and generalization in out-of-distribution environments. To leverage both paradigms, we propose Action-Draft-and-Verify (ADV): diffusion action expert drafts multiple candidate action chunks, and the VLM selects one by scoring all candidates in a single forward pass with a perplexity-style metric. Under matched backbones, training data, and action-chunk length, ADV improves success rate by +4.3 points in simulation and +19.7 points in real-world over diffusion-based baseline, with a single-pass VLM reranking overhead.

Latent actions serve as an intermediate representation that enables consistent modeling of vision-language-action (VLA) models across heterogeneous datasets. However, approaches to supervising VLAs with latent actions are fragmented and lack a systematic comparison. This work structures the study of latent action supervision from two perspectives: (i) regularizing the trajectory via image-based latent actions, and (ii) unifying the target space with action-based latent actions. Under a unified VLA baseline, we instantiate and compare four representative integration strategies. Our results reveal a formulation-task correspondence: image-based latent actions benefit long-horizon reasoning, whereas action-based latent actions excel at complex motor coordination. Furthermore, we find that directly supervising the VLM with discrete latent action tokens yields the most effective performance. Finally, our experiments offer initial insights into the benefits of latent action supervision in mixed-data, suggesting a promising direction for VLA training.

We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window. Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth. Together with Stable LatentMoE, which effectively activates 16 of 896 routed experts per token, and refined training and data recipes, these advances yield an approximately 2.5x improvement in overall scaling efficiency over Kimi K2. Post-training highlights reinforcement learning across general, agentic, and coding domains and multiple reasoning-effort levels, enabling compositional generalization and robust long-horizon execution. At 2.8T scale, Kimi K3 is supported by infrastructure advances in multiple areas: algorithm-system co-design for KDA, perfectly balanced expert-parallel training with efficient memory management, million-token agentic RL with persistent rollout and sandbox states, and deployment innovations. Extensive evaluations show that Kimi K3 achieves frontier-level performance across long-horizon coding, agentic, knowledge, reasoning, and vision tasks. While its overall performance still trails the most powerful proprietary models, namely Claude Fable 5 and GPT-5.6 Sol, Kimi K3 consistently outperforms other open and proprietary models evaluated in our suite. We release the full Kimi K3 model weights to facilitate future research and accelerate the broader deployment and adoption of frontier intelligence.

Multimodal Large Language Models (MLLMs) have shown remarkable progress in temporal or spatial localization tasks, but struggle with joint spatio-temporal video grounding (STVG). We identify two key bottlenecks hindering this capability: (1) the sheer number of visual tokens makes long-range and fine-grained visual modeling challenging; (2) generating a long sequence of bounding boxes in text makes it hard to accurately align each box with its specific video frame. Distinct from prior efforts that rely on attaching complex modules, we argue for a more elegant paradigm that unlocks the inherent potential of MLLMs and leverages their strengths. To this end, we propose \textbf{\textit{SpaceVLLM}}, a MLLM equipped with spatio-temporal video grounding capabilities. Specifically, we propose Spatio-Temporal Aware Queries, interleaved with video frames, to guide the MLLM in capturing both static appearance and dynamic motion features. We further present a lightweight Query-Guided Space Head that maps queries to precise spatial coordinates, bypassing the need for direct textual coordinate generation and enabling the MLLM to focus on video understanding. To further facilitate research in this area, we propose an automated data synthesis pipeline to construct \textbf{V-STG} dataset, comprising 110K STVG instances. Extensive experiments show that \textit{SpaceVLLM} achieves the state-of-the-art performance on STVG benchmarks and maintains strong performance on various video understanding tasks, validating our approach's effectiveness.

Cross-embodiment transfer in vision-language-action (VLA) models remains challenging because low-level state and action spaces differ fundamentally across robot platforms. We observe that the high-level cognitive process underlying manipulation, including scene perception, object identification, task planning, and sub-task decomposition, is largely shared across embodiments. Based on this observation, we present ZR-0, a 2.6 billion parameter end-to-end VLA model that uses dense Embodied Chain-of-Thought (ECoT) supervision to align cross-embodiment representations within the vision-language model (VLM). ZR-0 adopts a dual-stream architecture: a pre-trained VLM (System 2) generates structured ECoT reasoning during training, while a Diffusion Transformer-based action expert (System 1) produces continuous action chunks via flow matching. The two components are coupled through cross-attention, with an attention mask that restricts the action expert to input prompt features only, enabling ECoT generation to be entirely skipped at inference without any performance loss. ZR-0 is pre-trained on ProcCorpus-60M, a large-scale dataset comprising approximately 60 million frames (approximately 1,000 hours) from over 400K trajectories, with dense ECoT annotations covering 96.8

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× 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.

Large language models (LLMs) have exhibited remarkable performance on complex reasoning tasks, with reinforcement learning under verifiable rewards (RLVR) emerging as a principled framework for aligning model behavior with reasoning chains. Despite its promise, RLVR remains prohibitively resource-intensive, requiring extensive reward signals and incurring substantial rollout costs during training. In this work, we revisit the fundamental question of data and compute efficiency in RLVR. We first establish a theoretical lower bound on the sample complexity required to unlock reasoning capabilities, and empirically validate that strong performance can be achieved with a surprisingly small number of training instances. To tackle the computational burden, we propose Dynamic One-Shot Policy Refinement (DoPR), a uncertainty-aware RL strategy that dynamically selects a single informative training sample per batch for policy updates, guided by reward volatility and exploration-driven acquisition. DoPR reduces rollout overhead by nearly an order of magnitude while preserving competitive reasoning accuracy, offering a scalable and resource-efficient solution for LLM post-training. This approach offers a practical path toward more efficient and accessible RL-based training for reasoning-intensive LLM applications.

Semantic communication is a new paradigm that aims at providing more efficient communication for the next-generation wireless network. It focuses on transmitting extracted, meaningful information instead of the raw data. However, deep learning-enabled image semantic communication models often require a significant amount of time and energy for training, which is unacceptable, especially for mobile devices. To solve this challenge, our paper first introduces a distributed image semantic communication system where the base station and local devices will collaboratively train the models for uplink communication. Furthermore, we formulate a joint optimization problem to balance time and energy consumption on the local devices during training while ensuring effective model performance. An adaptable resource allocation algorithm is proposed to meet requirements under different scenarios, and its time complexity, solution quality, and convergence are thoroughly analyzed. Experimental results demonstrate the superiority of our algorithm in resource allocation optimization against existing benchmarks and discuss its impact on the performance of image semantic communication systems.

Economic development hinges on structural change, that is, transformations in what an economy produces. The field of economic complexity has investigated this process through two related but distinct branches: one studying how economies diversify, the other how the complexity of an economy is reflected in its output. However, a formal connection between these approaches, and their relationship to classic accounts of structural transformation (for example, from agriculture to manufacturing), remains unclear. Here we introduce a simple dynamical model that links these perspectives through one core idea: economies diversify preferentially into activities related to those they already do. Studying this model yields three main results: It generates quantities resembling economic complexity metrics, suggests these metrics summarize long-term structural change rather than directly infer an economy's complexity, and reproduces stylized facts of development. Our framework formally connects the field's conceptual strands, bridges short and long timescales of change, and adds granularity to classic descriptions of development.

Description-based text-to-speech (TTS) models exhibit strong performance on in-domain text descriptions, i.e., those encountered during training. However, in real-world applications, the diverse range of user-generated descriptions inevitably introduces numerous out-of-domain inputs that challenge the text understanding capabilities of these systems. To address this issue, we propose MoE-TTS, a description-based TTS model designed to enhance the understanding of out-of-domain text descriptions. MoE-TTS employs a modality-based mixture-of-experts (MoE) approach to augment a pre-trained textual large language model (LLM) with a set of specialized weights adapted to the speech modality while maintaining the original LLM frozen during training. This approach allows MoE-TTS to effectively leverage the pre-trained knowledge and text understanding abilities of textual LLMs. Our experimental results indicate that: first, even the most advanced closed-source commercial products can be challenged by carefully designed out-of-domain description test sets; second, MoE-TTS achieves superior performance in generating speech that more accurately reflects the descriptions. We encourage readers to listen to the demos at https://welkinyang.github.io/MoE-TTS/.

The construction sector, as the prime culprit that aggravates greenhouse gas emissions, has initiated transitions in construction methods for emissions abatement. In such a context, prefabricated construction has been widely promoted in developing countries. This paper assesses the effects of three environmental policies: carbon tax, green subsidy, and unit subsidy on carbon allocation and emissions abatement of prefabricated construction supply chain (PCSC) consisting of a prefab manufacturer (PM) and a project developer (PD). In our game model, the PD determines a prefabricated ratio, the ratio of prefabs to the total construction components, and purchases prefabs from the PM for assembly construction, which enables fewer emissions than on-site construction. We find that the carbon tax has a compound pass-through effect in PCSC, which imposes a quasi-subsidy effect on the PD, allowing the PM to achieve a prefab premium and thus inducing more intense double marginalization than that under the green subsidy. The unit subsidy holds a cost-compensating effect, which mitigates double marginalization arising from the spill-over effect of the green subsidy. Nevertheless, both subsidies lead to identical carbon allocation and emissions abatement of PCSC. The carbon tax outperforms subsidy policies in reducing total emissions and emission intensity, and is superior in responsibility-benefit balance in PCSC. Our study offers insights into the government's choice of environmental policies and the sustainable production decisions of PCSC members while, theoretically, contributing to the sustainable supply chain theory by revealing the impacts of environmental policies on the supply chain for transitional green products, such as PCSC.

Trajectory generation has recently drawn growing interest in privacy-preserving urban mobility studies and location-based service applications. Although many studies have used deep learning or generative AI methods to model trajectories and have achieved promising results, the robustness and interpretability of such models are largely unexplored. This limits the application of trajectory generation algorithms on noisy real-world data and their trustworthiness in downstream tasks. To address this issue, we exploit the regular structure in urban trajectories and propose a deep generative model based on the pathlet representation, which encode trajectories with binary vectors associated with a learned dictionary of trajectory segments. Specifically, we introduce a probabilistic graphical model to describe the trajectory generation process, which includes a Variational Autoencoder (VAE) component and a linear decoder component. During training, the model can simultaneously learn the latent embedding of pathlet representations and the pathlet dictionary that captures mobility patterns in the trajectory dataset. The conditional version of our model can also be used to generate customized trajectories based on temporal and spatial constraints. Our model can effectively learn data distribution even using noisy data, achieving relative improvements of 35.4% and 26.3% over strong baselines on two real-world trajectory datasets. Moreover, the generated trajectories can be conveniently utilized for multiple downstream tasks, including trajectory prediction and data denoising. Lastly, the framework design offers a significant efficiency advantage, saving 64.8% of the time and 56.5% of GPU memory compared to previous approaches.

This paper examines how carbon emissions affect municipal corporate bond (MCB) pricing in China. Using data from 2008-2022, we find that higher emissions lead to wider credit spreads. To address endogeneity, we use a difference-in-differences strategy and an instrumental variable approach, both confirming the results. Mechanism analysis shows that carbon emissions widen credit spreads by weakening implicit guarantees, reducing repayment capacity, and lowering investor demand. The effect is stronger in developed cities, those with limited refinancing capacity, lower marketization, lower fiscal transparency, and after 2020. These findings highlight the financial implications of environmental risk in municipal bond markets.

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.

Low-carbon district heating systems are the result of the improvement of existing fossil-based district heating systems, where various environmentally friendly heat sources are integrated to decarbonize the heating sector. Existing research has the problem of emphasizing "quantity" over "quality" in the planning stage of the district heating system and does not consider the load elasticity of the demand side, resulting in redundancy in the equipment capacity. Therefore, a source-load-storage collaborative planning approach for a low-carbon district heating system is proposed, which comprehensively considers the grade equivalence of thermal energy supply and demand, and integrated electricity-heat demand response to achieve the optimal system configuration. Additionally, we establish a refined mathematical model of latent heat thermal energy storage based on engineering-measured data to address the issue that a general energy storage planning model cannot accurately describe the differentiated charging-discharging characteristics of multi-energy storage systems. A case study is considered, and the results are thoroughly analyzed. Compared to the conventional planning method, the proposed method can effectively reduce the capacity of system equipment by up to 35.0% and reduce the annual investment cost by approximately 5.31%. The outcomes can be referred to for district heating system planning decisions.

Accurate daily predictions of industrial carbon emissions can improve our understanding of industrial activities. In this paper, we propose a novel approach to predict daily industrial carbon emissions by fine-tuning pre-trained language models. This method constructs a general carbon emissions model applicable to all provinces, providing accurate predictions in natural language format and outperforming traditional carbon emissions models. It achieves Mean Absolute Error (MAE) of 0.005757, Root Mean Squared Error (RMSE) of 0.008982, and Mean Absolute Percentage Error (MAPE) of 5.485714 %. The results suggest that fine-tuned T5 with prompt text significantly improves the predictive accuracy for daily industrial carbon emissions compared with typical multi- layer perception and long short-term memory numerical models. A single model can be adapted to the prediction of industrial carbon emissions in 31 provinces of China.

Achieving net-zero emissions requires rapid innovation, yet the necessary technological knowhow is scattered across industries and countries. Comparing functionally similar green and nongreen patents, we identify "Green Building Blocks" (GBBs): modular components that can be added to reduce existing technologies' carbon footprints. These GBBs depict the anatomy of the green transition as a network that connects problems – nongreen technologies – to GBBs that mitigate their climate-change impact. Node degrees in this network are highly unequal, showing that the scope for climate-change mitigating innovation varies substantially across domains. The network also helps predict which green technologies firms develop themselves, and which alliances they form to do so. This reveals a critical dependence on international collaboration: optimal innovation partners for 84

Climate policies that target carbon emissions can induce co-benefits for air quality. Previous urban studies have typically focused on either carbon reduction or air pollution control independently, but few have examined their combined effects on reducing carbon emissions and consequential environmental gains. We develop an integrated modelling framework to assess the impacts of different low-carbon transitions and end-of-pipe controls on PM2.5 and ozone concentrations and associated premature mortality in the megacity of Guangzhou. The results show that the implementation of both deep carbon mitigation and aggressive air pollution control policies can reduce the city's pollutant emissions to 34%-51% of the 2020 levels by 2035. Consequently, the population-weighted PM2.5 concentration in 2035 is projected to decrease by 5 mg/m3 compared to the 2035 baseline scenario. However, the ozone concentration is expected to rise by 35 mg/m3 due to the reduced titration effect of NO on ozone. These changes are estimated to prevent approximately 3.0 thousand (95% CI: 2.0-3.9) PM2.5-related premature deaths, while increasing ozone-related premature deaths by approximately 1.6 thousand (95% CI: 0.7-2.7). Moreover, implementing multiregional integrated control measures in Guangzhou and its neighbouring cities yields greater air quality and health benefits for Guangzhou compared to local enforcement alone, resulting in 1.5 times more avoided PM2.5-related premature deaths. Additionally, the increase in ozone-related premature deaths from these cooperative emission control strategies is merely 0.3 times the figure observed under local enforcement alone. The transport and industry sectors play a crucial role in reducing air pollutant emissions, whereas reductions in the solvent use sector can help mitigate the adverse effects of reduced NOx on ozone pollution. These findings highlight the need for comprehensively multiregional strategies to balance the trade-offs between reducing PM2.5 and ozone-related health impacts, offering valuable insights for urban policy makers aiming to optimize both climate and air quality goals on a broader scale.

The impact of disturbances on a transportation network varies depending on the location and characteristics of the affected highway segments. Given limited resources, it is crucial to prioritize the protection and repair of highway segments based on their importance to maintaining overall network performance during disruptions. This paper proposes a novel method for ranking the importance of highway segments, leveraging a novel local–transit percolation and clustering-based method. Initially, the highway network is constructed by Graph theory, and the k-means clustering method is applied considering each segment’s transit and local traffic flows. Subsequently, a local–transit percolation model is constructed to generate an initial ranking of segments based on the size of the second-largest clusters during the percolation phase transition. A secondary ranking is performed by refining the results from the clustering phase. Results of a control experiment show that, compared to baselines, the proposed ranking approach demonstrates a significantly improved ability to sustain network demand and connectivity when high-ranked segments are moved. The model uncertainty analysis was conducted by adding noise to the gantry records, and the experiments demonstrated that the model exhibits robustness under noisy conditions. These findings highlight the effectiveness and superiority of the proposed method.

We introduce TableLLM, a robust large language model (LLM) with 8 billion parameters, purpose-built for proficiently handling tabular data manipulation tasks, whether they are embedded within documents or spreadsheets, catering to real-world office scenarios. We propose a distant supervision method for training, which comprises a reasoning process extension strategy, aiding in training LLMs to understand reasoning patterns more effectively as well as a cross-way validation strategy, ensuring the quality of the automatically generated data. To evaluate the performance of TableLLM, we have crafted benchmarks tailored to address both document and spreadsheet formats as well as constructed a well-organized evaluation pipeline capable of handling both scenarios. Thorough evaluations underscore the advantages of TableLLM when compared to various existing general-purpose and tabular data-focused LLMs. We have publicly released the model checkpoint, source code, benchmarks, and a web application for user interaction. Our codes and data are publicly available at https://github.com/TableLLM/TableLLM.