Haotian Zhao

Jiawei Wang, Qinglai Guo, Haotian Zhao et al.

The interaction between extreme weather events and interdependent critical infrastructure systems involves complex spatiotemporal dynamics. Multi-type emergency decisions within energy-transportation infrastructures significantly influence system performance throughout the extreme weather process. A comprehensive assessment of these factors faces challenges in model complexity, heterogeneous differences between energy and transportation systems, and cross-sector privacy. This paper proposes a risk assessment framework that integrates the heterogeneous energy and transportation systems in the form of a unified network flow model, which enables full accommodation of multiple types of energy-transportation emergency decisions while capturing the compound spatiotemporal impacts of extreme weather on both systems simultaneously. Based on this framework, a targeted method for identifying system vulnerabilities is further developed. This method employs neural network surrogates to achieve privacy protection and accelerated identification while maintaining consideration of system interdependencies. Numerical experiments demonstrate that the proposed framework and method can reveal the risk levels faced by urban infrastructure systems, identify vulnerabilities that should be prioritized for reinforcement, and strike a balance between accuracy and speed.

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.

Wenyu Zhang, Guoliang You, Tianlun et al.

Coding agents are increasingly used as application builders, yet many evaluations still focus on source code, repository-level tests, or intermediate traces rather than the delivered application. We introduce WebGameBench, a requirement-to-application benchmark that evaluates whether coding agents can turn a frozen Structured WebGame Specification into a browser-accessible game. Browser-native games provide a compact but behavior-dense testbed: even simple games require coordinated input handling, spatial mapping, rule execution, state transitions, terminal conditions, restart behavior, and visible feedback. In WebGameBench, each generated artifact is built, served, and exposed as a browser-accessible application under a unified deployment protocol. A runtime evaluator then interacts with the delivered game in a real browser and assigns a three-way label: EXCELLENT, USABLE, or UNUSABLE. On a human-reviewed subset, the runtime label is broadly aligned with human gameplay review under the Usable-rate criterion. Across 111 tasks, 12 coding agents, and 14 evaluation configurations, WebGameBench separates current systems: the best configuration reaches a 76.9% usable rate but only a 20.2% excellent rate. This gap shows that crossing the minimum playable-delivery threshold is still far from complete requirement satisfaction. To our knowledge, WebGameBench is the first requirement-to-application benchmark for browser-native game delivery that validates delivered-application runtime labels against independent human gameplay review under the Usable-rate criterion.

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%).

Qinglong Yang, Haoming Li, Haotian Zhao et al.

Mobile GUI agents are becoming critical tools for enhancing human-device interaction efficiency, with multimodal large language models (MLLMs) emerging as dominant paradigms in this domain. Current agents, however, are limited to following explicit human instructions, resulting in insufficient capability for proactive intent anticipation. Additionally, these agents fail to leverage the contextual information associated with users during task execution, thereby neglecting potentially vast differences in user preferences. To address these challenges, we introduce the FingerTip benchmark. It contains two new tracks: proactive task suggestions by analyzing environment observation and users' previous intents, and personalized task execution by catering to users' action preferences. We collected unique human demonstrations of multi-step Android device interactions across a variety of everyday apps. These demonstrations are not isolated but are continuously acquired from the users' long-term usage in their real lives, and encompass essential user-related contextual information. Our experiments reveal challenges of the tasks we propose. The model fine-tuned with the data we collected effectively utilized user information and achieved good results, highlighting the potential of our approach in building more user-oriented mobile GUI agents. Our code is open-source at https://anonymous.4open.science/r/FingerTip-57B8 for reproducibility.

Tianshu Zhu, Wenyu Zhang, Xiaoying Zuo et al.

SWE-bench-style agentic reinforcement learning relies on expensive stateful trajectories, yet substantial compute is wasted on sampled rollout groups with skewed pass rates, where binary rewards provide a weak contrastive signal. We frame this inefficiency as a pass-rate control problem and show that a 50% pass rate is the most informative operating point: it maximizes reward entropy, the probability of surviving group filtering, RLOO advantage energy under GRPO, and success--failure contrastive structure. Guided by this principle, we propose Prefix Sampling (PS), which replays trajectory prefixes to steer skewed groups toward this regime: successful prefixes serve as head starts for mostly failing groups, while failing prefixes serve as handicaps for mostly passing groups. In stateful agent environments, prefix states are reconstructed through replay while replayed tokens are excluded from the loss, restricting optimization to continuations generated by the current policy. On SWE-bench-style agentic RL, PS delivers end-to-end wall-clock speedups of 2.01x on Qwen3-14B and 1.55x on Qwen3-32B while preserving or improving final verified performance. For 14B, the SWE-bench Verified peak rises from the baseline peak of 0.273 to 0.295 under PS. Additional mathematical reasoning experiments on AIME 2025 show the same pass-rate control pattern and decompose the gains into replay, bidirectional coverage, and adaptive control.

Haotian Zhao, Yuxin Zhang, Songlin Zhou et al.

Reinforcement learning (RL) has significantly advanced the ability of large language model (LLM) agents to interact with environments and solve multi-turn tasks. Yet effective training remains challenging, as sparse, outcome-only rewards make it difficult to assign credit to individual steps in an agent's action trajectory. A common remedy is to introduce dense intermediate supervision, such as process reward models or auxiliary self-supervised signals, but this increases supervision and tuning complexity and often generalizes poorly across tasks and domains. This paper presents AEM, a supervision-free credit assignment method that adaptively modulates entropy dynamics during RL training to achieve a more effective exploration-exploitation trade-off. Theoretically, we elevate entropy analysis from the token level to the response level to reduce token sampling variance and show that entropy drift under natural gradients is intrinsically governed by the product of the advantage and the relative response surprisal. Specifically, we derive a practical proxy to reshape training dynamics, enabling a natural transition from exploration to exploitation. Extensive experiments across various benchmarks and models ranging from 1.5B to 32B parameters demonstrate the effectiveness of AEM, including a notable 1.4 percent gain when integrated into a state-of-the-art baseline on the highly challenging SWE-bench-Verified benchmark.

Haotian Zhao, Ruifeng Chen, Jing Yan et al.

To shorten the door-to-puncture time for better treating patients with acute ischemic stroke, it is highly desired to obtain quantitative cerebral perfusion images using C-arm cone-beam computed tomography (CBCT) equipped in the interventional suite. However, limited by the slow gantry rotation speed, the temporal resolution and temporal sampling density of typical C-arm CBCT are much poorer than those of multi-detector-row CT in the diagnostic imaging suite. The current quantitative perfusion imaging includes two cascaded steps: time-resolved image reconstruction and perfusion parametric estimation. For time-resolved image reconstruction, the technical challenge imposed by poor temporal resolution and poor sampling density causes inaccurate quantification of the temporal variation of cerebral artery and tissue attenuation values. For perfusion parametric estimation, it remains a technical challenge to appropriately design the handcrafted regularization for better solving the associated deconvolution problem. These two challenges together prevent obtaining quantitatively accurate perfusion images using C-arm CBCT. The purpose of this work is to simultaneously address these two challenges by combining the two cascaded steps into a single joint optimization problem and reconstructing quantitative perfusion images directly from the measured sinogram data. In the developed direct cerebral perfusion parametric image reconstruction technique, TRAINER in short, the quantitative perfusion images have been represented as a subject-specific conditional generative model trained under the constraint of the time-resolved CT forward model, perfusion convolutional model, and the subject's own measured sinogram data. Results shown in this paper demonstrated that using TRAINER, quantitative cerebral perfusion images can be accurately obtained using C-arm CBCT in the interventional suite.