GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

We present GLM-4.5, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes. Through multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning, GLM-4.5 achieves strong performance across agentic, reasoning, and coding (ARC) tasks, scoring 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified. With much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks. We release both GLM-4.5 (355B parameters) and a compact version, GLM-4.5-Air (106B parameters), to advance research in reasoning and agentic AI systems. Code, models, and more information are available at https://github.com/zai-org/GLM-4.5.

ExAnte: A Benchmark for Ex-Ante Inference in Large Language Models

Large language models (LLMs) face significant challenges in ex-ante reasoning, where analysis, inference, or predictions must be made without access to information from future events. Even with explicit prompts enforcing temporal cutoffs, LLMs often generate outputs influenced by internalized knowledge of events beyond the specified cutoff. This paper introduces a novel task and benchmark designed to evaluate the ability of LLMs to reason while adhering to such temporal constraints. The benchmark includes a variety of tasks: stock prediction, Wikipedia event prediction, scientific publication prediction, and Question Answering (QA), designed to assess factual knowledge under temporal cutoff constraints. We use leakage rate to quantify models' reliance on future information beyond cutoff timestamps. Experimental results reveal that LLMs struggle to consistently adhere to temporal cutoffs across common prompting strategies and tasks, demonstrating persistent challenges in ex-ante reasoning. This benchmark provides a potential evaluation framework to advance the development of LLMs' temporal reasoning ability for time-sensitive applications.

NurValues: Real-World Nursing Values Evaluation for Large Language Models in Clinical Context

This work introduces the first benchmark for nursing value alignment, consisting of five core value dimensions distilled from international nursing codes: Altruism, Human Dignity, Integrity, Justice, and Professionalism. The benchmark comprises 1,100 real-world nursing behavior instances collected through a five-month longitudinal field study across three hospitals of varying tiers. These instances are annotated by five clinical nurses and then augmented with LLM-generated counterfactuals with reversed ethic polarity. Each original case is paired with a value-aligned and a value-violating version, resulting in 2,200 labeled instances that constitute the Easy-Level dataset. To increase adversarial complexity, each instance is further transformed into a dialogue-based format that embeds contextual cues and subtle misleading signals, yielding a Hard-Level dataset. We evaluate 23 state-of-the-art (SoTA) LLMs on their alignment with nursing values. Our findings reveal three key insights: (1) DeepSeek-V3 achieves the highest performance on the Easy-Level dataset (94.55), where Claude 3.5 Sonnet outperforms other models on the Hard-Level dataset (89.43), significantly surpassing the medical LLMs; (2) Justice is consistently the most difficult nursing value dimension to evaluate; and (3) in-context learning significantly improves alignment. This work aims to provide a foundation for value-sensitive LLMs development in clinical settings. The dataset and the code are available at https://huggingface.co/datasets/Ben012345/NurValues.

PHYBench: Holistic Evaluation of Physical Perception and Reasoning in Large Language Models

We introduce PHYBench, a novel, high-quality benchmark designed for evaluating reasoning capabilities of large language models (LLMs) in physical contexts. PHYBench consists of 500 meticulously curated physics problems based on real-world physical scenarios, designed to assess the ability of models to understand and reason about realistic physical processes. Covering mechanics, electromagnetism, thermodynamics, optics, modern physics, and advanced physics, the benchmark spans difficulty levels from high school exercises to undergraduate problems and Physics Olympiad challenges. Additionally, we propose the Expression Edit Distance (EED) Score, a novel evaluation metric based on the edit distance between mathematical expressions, which effectively captures differences in model reasoning processes and results beyond traditional binary scoring methods. We evaluate various LLMs on PHYBench and compare their performance with human experts. Our results reveal that even state-of-the-art reasoning models significantly lag behind human experts, highlighting their limitations and the need for improvement in complex physical reasoning scenarios. Our benchmark results and dataset are publicly available at https://phybench-official.github.io/phybench-demo/.

FairACE: Achieving Degree Fairness in Graph Neural Networks via Contrastive and Adversarial Group-Balanced Training

Fairness has been a significant challenge in graph neural networks (GNNs) since degree biases often result in un-equal prediction performance among nodes with varying degrees. Existing GNN models focus on prediction accuracy, frequently overlooking fairness across different degree groups. To addressthis issue, we propose a novel GNN framework, namely Fairness- Aware Asymmetric Contrastive Ensemble (FairACE), which inte-grates asymmetric contrastive learning with adversarial training to improve degree fairness. FairACE captures one-hop local neighborhood information and two-hop monophily similarity to create fairer node representations and employs a degree fairness regulator to balance performance between high-degree and low-degree nodes. During model training, a novel group-balanced fairness loss is proposed to minimize classification disparities across degree groups. In addition, we also propose a novel fairness metric, the Accuracy Distribution Gap (ADG), which can quantitatively assess and ensure equitable performance across different degree-based node groups. Experimental results on both synthetic and real-world datasets demonstrate that FairACE significantly improves degree fairness metrics while maintaining competitive accuracy in comparison to the state-of-the-art GNN models.

Policy Learning with a Natural Language Action Space: A Causal Approach

This paper introduces a novel causal framework for multi-stage decision-making in natural language action spaces where outcomes are only observed after a sequence of actions. While recent approaches like Proximal Policy Optimization (PPO) can handle such delayed-reward settings in high-dimensional action spaces, they typically require multiple models (policy, value, and reward) and substantial training data. Our approach employs Q-learning to estimate Dynamic Treatment Regimes (DTR) through a single model, enabling data-efficient policy learning via gradient ascent on language embeddings. A key technical contribution of our approach is a decoding strategy that translates optimized embeddings back into coherent natural language. We evaluate our approach on mental health intervention, hate speech countering, and sentiment transfer tasks, demonstrating significant improvements over competitive baselines across multiple metrics. Notably, our method achieves superior transfer strength while maintaining content preservation and fluency, as validated through human evaluation. Our work provides a practical foundation for learning optimal policies in complex language tasks where training data is limited.

CoT-based Synthesizer: Enhancing LLM Performance through Answer Synthesis

Current inference scaling methods, such as Self-consistency and Best-of-N, have proven effective in improving the accuracy of LLMs on complex reasoning tasks. However, these methods rely heavily on the quality of candidate responses and are unable to produce correct answers when all candidates are incorrect. In this paper, we propose a novel inference scaling strategy, CoT-based Synthesizer, which leverages CoT reasoning to synthesize superior answers by analyzing complementary information from multiple candidate responses, even when all candidate responses are flawed. To enable a lightweight and cost-effective implementation, we introduce an automated data generation pipeline that creates diverse training data. This allows smaller LLMs trained on this data to improve the inference accuracy of larger models, including API-based LLMs. Experimental results across four benchmark datasets with seven policy models demonstrate that our method significantly enhances performance, with gains of 11.8% for Llama3-8B and 10.3% for GPT-4o on the MATH dataset. The corresponding training data and code are publicly available on https://github.com/RUCKBReasoning/CoT-based-Synthesizer.

Metasurface-generated large and arbitrary analog convolution kernels for accelerated machine vision

In the rapidly evolving field of artificial intelligence, convolutional neural networks are essential for tackling complex challenges such as machine vision and medical diagnosis. Recently, to address the challenges in processing speed and power consumption of conventional digital convolution operations, many optical components have been suggested to replace the digital convolution layer in the neural network, accelerating various machine vision tasks. Nonetheless, the analog nature of the optical convolution kernel has not been fully explored. Here, we develop a spatial frequency domain training method to create arbitrarily shaped analog convolution kernels using an optical metasurface as the convolution layer, with its receptive field largely surpassing digital convolution kernels. By employing spatial multiplexing, the multiple parallel convolution kernels with both positive and negative weights are generated under the incoherent illumination condition. We experimentally demonstrate a 98.59% classification accuracy on the MNIST dataset, with simulations showing 92.63% and 68.67% accuracy on the Fashion-MNIST and CIFAR-10 datasets with additional digital layers. This work underscores the unique advantage of analog optical convolution, offering a promising avenue to accelerate machine vision tasks, especially in edge devices.

DroneWiS: Automated Simulation Testing of small Unmanned Aerial Systems in Realistic Windy Conditions

The continuous evolution of small Unmanned Aerial Systems (sUAS) demands advanced testing methodologies to ensure their safe and reliable operations in the real-world. To push the boundaries of sUAS simulation testing in realistic environments, we previously developed the DroneReqValidator (DRV) platform, allowing developers to automatically conduct simulation testing in digital twin of earth. In this paper, we present DRV 2.0, which introduces a novel component called DroneWiS (Drone Wind Simulation). DroneWiS allows sUAS developers to automatically simulate realistic windy conditions and test the resilience of sUAS against wind. Unlike current state-of-the-art simulation tools such as Gazebo and AirSim that only simulate basic wind conditions, DroneWiS leverages Computational Fluid Dynamics (CFD) to compute the unique wind flows caused by the interaction of wind with the objects in the environment such as buildings and uneven terrains. This simulation capability provides deeper insights to developers about the navigation capability of sUAS in challenging and realistic windy conditions. DroneWiS equips sUAS developers with a powerful tool to test, debug, and improve the reliability and safety of sUAS in real-world. A working demonstration is available at https://youtu.be/khBHEBST8Wc

SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulation

We introduce SpreadsheetBench, a challenging spreadsheet manipulation benchmark exclusively derived from real-world scenarios, designed to immerse current large language models (LLMs) in the actual workflow of spreadsheet users. Unlike existing benchmarks that rely on synthesized queries and simplified spreadsheet files, SpreadsheetBench is built from 912 real questions gathered from online Excel forums, which reflect the intricate needs of users. The associated spreadsheets from the forums contain a variety of tabular data such as multiple tables, non-standard relational tables, and abundant non-textual elements. Furthermore, we propose a more reliable evaluation metric akin to online judge platforms, where multiple spreadsheet files are created as test cases for each instruction, ensuring the evaluation of robust solutions capable of handling spreadsheets with varying values. Our comprehensive evaluation of various LLMs under both single-round and multi-round inference settings reveals a substantial gap between the state-of-the-art (SOTA) models and human performance, highlighting the benchmark's difficulty.