The paper introduces a multi‑metric benchmark for evaluating spatial climate downscaling methods, comparing five techniques on ERA5 temperature, wind, and precipitation data. Five complementary criteria—pointwise error, structural similarity, distribution error, spectral error, and gradient error—are used to assess each method’s performance. Results show systematic trade‑offs: some methods excel in pointwise and spatially aligned metrics but lose high‑frequency content, while others preserve more spectral variability at the expense of local structure accuracy, leading to variable method rankings across metrics and variables.
By Loys Masquelier, Etienne Le Naour
Hob‑VL is a benchmark for visually grounded Boolean reasoning that tests models on two tasks: determining whether a Boolean rule holds in an image and identifying the unique object that satisfies a Boolean description. It contains 6,000 balanced Yes/No questions built from ten visual statements across 1,000 generated scenes and 46 labeled photographs, plus 1,000 object‑identification questions on the same photographs. The questions are designed to challenge reasoning with misleading local cues, nested logical operations, and both symbolic and natural‑language presentations, revealing significant gaps in current model performance.
By Yuzhou Wang, Emile Anand, Ijay Narang
The paper introduces the Separable Law, a framework that predicts how vision‑language model performance varies with language backbone size and visual token count. By fitting this law to 26 InternVL and QwenVL models across a range of backbone sizes (1B–72B) and image resolutions (224–8K pixels), the authors show that some question types scale predictably with model capacity while others do not. The law also provides a closed‑form rule for allocating compute between backbone size and visual tokens, helping to choose near‑optimal model and image sizes under a fixed budget.
By Xinye Zhao, Yunkai Dang, Yunchen Wu, Wenbin Li
Architectural Sampling is a training‑free technique that improves test‑time scaling for frozen vision‑language models by generating diverse candidates through distinct forward computations. It achieves this by reusing selected blocks of decoder layers, varying block location and repetition count, thereby creating computational diversity without updating weights or adding parameters. Experiments on five Qwen checkpoints across twelve multimodal benchmarks show that this method raises pass@9 by an average of 6.58 percentage points over standard temperature sampling, with early‑layer reuse delivering the strongest gains and lower lexical overlap in the generated candidates.
By Akshit Singh, Shyam Marjit, Wei Lin, Leonid Karlinsky, M. Jehanzeb Mirza
The paper presents an analysis of asynchronous reinforcement learning for large language model post‑training, focusing on the impact of stale rollouts. It derives a convergence bound for GRPO‑style algorithms that balances the gradient estimator’s second moment and bias, and introduces a new group‑mass capping GRPO (GMC‑GRPO) method that reduces delay‑dependent bias. Experiments on Qwen3 models and reasoning benchmarks show that GMC‑GRPO outperforms existing baselines in robustness to large rollout delays.
By Qijia He, Ruinan Jin, Jun Luo, Shaofeng Zou, Yingbin Liang
DuoMind is a distributed hierarchical framework that enables multi-robot coordination by combining vision-language-action models for low‑level execution with vision-language models for high‑level reasoning and inter‑robot communication. Each robot’s orchestrator processes task instructions, local observations, and messages from peers to generate precise action commands and semantic messages for other robots. The authors introduce RoboPoly, a benchmark of long‑horizon manipulation tasks, and show through experiments that DuoMind improves multi‑robot task performance, with ablation studies highlighting the roles of hierarchical orchestration and semantic communication.
By Hanchu Zhou, Dechen Gao, Hang Wang, Brendan Lynch, Boqi Zhao, Qiyao Ma, Raman Goyal, Junshan Zhang
KaliBench is a fine‑grained benchmark that evaluates large language models’ ability to translate natural‑language queries into executable command‑line instructions for Kali Linux. It contains 8,504 query–command pairs covering 1,642 tools across 23 capability dimensions and 5 security phases, and uses a deterministic, alias‑aware evaluation pipeline with multi‑stage verification (LLM validation, sandboxed execution, and human refinement). The benchmark also supports runtime‑free verifiable rewards, enabling training improvements that bring smaller models close to the performance of a 685B MoE model.
By Pengfei Li, Naufal Suryanto, Sicheng Zhang, Muzammal Naseer
AstroAgentBench is a seven-family benchmark designed to evaluate agentic planning for space mission tasks such as scheduling, observation planning, constellation design, and relay support. The benchmark requires agents to submit planning artifacts that are verified for schema, timing, geometry, resources, and mission value, with results reported as validity and normalized scores compared to task-specific solver references. Across five LLM agent systems and 35 held-out cases, the strongest systems approach or exceed solver-reference scores on several families, while weaker systems often fail to produce high-value valid plans, and even strong systems lose quality on geometric, product-level, or design-heavy tasks. Failure analyses identify task-contract misformulation and weak solution construction as key issues, and successful runs demonstrate that agent calibration against verifier feedback and adaptive search can improve performance; ablation studies show that procedure injection and memory accumulation help when they provide the missing formulation, calibration, or search support.
By Weiyi Wang, Xinchi Chen, Jingjing Gong, Xuanjing Huang, Xipeng Qiu
HyperGuide introduces a hyperbolic geometric signal to guide multi-step reasoning in large language models, addressing the trade-off between efficient single-pass generation and computationally heavy tree-search methods. By projecting LLM hidden states into hyperbolic space, the approach leverages the space’s asymmetry—compact near the origin and exponentially expanding toward the boundary—to encode solution proximity and branch differentiation. A lightweight head and a fine-tuned low-rank adapter use this signal to improve reasoning accuracy, especially on deeper reasoning chains, with consistent gains across multiple benchmarks.
By Yuyu Liu, Haotian Xu, Yanan He, Sarang Rajendra Patil, Mengjia Xu, Tengfei Ma
The paper introduces IMAVB, a 500‑clip benchmark that tests whether omnimodal large language models can detect when a textual premise contradicts their visual or audio input. Experiments on eight open‑source models and Gemini 3.1 Pro reveal a Representation‑Action Gap: internal states encode mismatches, yet the models rarely reject false premises, exhibiting under‑rejection or over‑rejection. A probe‑guided logit adjustment improves rejection behavior, suggesting the main bottleneck is in translating perception to action rather than in perception itself.
By Trung Nguyen Quang, Yiming Gao, Fanyi Pu, Kaichen Zhang, Shuo Sun, Ziwei Liu
The paper introduces masked self‑distillation, a post‑training framework that trains a language model to internalize portions of its own intermediate reasoning traces. By varying the fraction of trace internalized, the authors demonstrate that models can achieve higher inference efficiency and improved task performance on math and graph‑coloring problems. Experiments on Qwen3‑4B and Qwen3‑8B show that the method generalizes well to in‑domain out‑of‑distribution cases without catastrophic forgetting, and that supervised fine‑tuning alone can reduce trace length at the expense of generalization.
By Durgesh Kalwar, Vardhan Palod, Jaya Adithya Pavuluri, Subbarao Kambhampati
CANTO is a transformer neural operator that directly maps continuous CAD geometry—represented as sequences of NURBS patches—to physical fields without requiring meshing. It tokenizes control points, knot vectors, and weights to predict continuous surface and volume fields at arbitrary query locations. Evaluated on automotive and aircraft aerodynamics benchmarks, CANTO achieves state‑of‑the‑art accuracy, notably a 19.8% reduction in surface‑pressure error on HiLiftAeroML, and enables gradient‑based inverse design, finding lower‑drag designs on AhmedML.
By Daniel Leibovici, Nikola Borislavov Kovachki, Dawon Ahn, Ruben Ohana, Ira J. S. Shokar, Abouzar Ghasemi, Semih Akkurt, Rishikesh Ranade, Neil Ashton, Jan Kautz, Jean Kossaifi
FuncBridge is a two‑stage framework that addresses functional generalization in robotics by decoupling functional reasoning from action execution. It learns to predict generalizable 2D keypoint trajectories from action‑free data and then grounds these trajectories into robot actions with limited demonstrations. Across a benchmark of ten tools and three functions—hitting, sweeping, and hooking—FuncBridge outperforms state‑of‑the‑art methods on unseen tools in both simulation and real‑world tests.
By Chuhao Zhou, Liquan Wang, Shuxin Cao, Xiangyu Chen, Yuxuan Hu, Boyu Ma, Animesh Garg, Jianfei Yang
Pepti-drift is a one‑step peptide generation framework that uses a single latent refinement and parallel decoding, avoiding costly inference‑time guidance. It incorporates attraction toward target‑specific binders and repulsion from liability‑associated regions, achieving an 18.37% predicted Safe‑Active yield across 88 held‑out targets. The method delivers competitive gains to multi‑property‑guided baselines while reducing generation cost by 468×, enabling scalable and fair high‑throughput peptide design.
By Takashi Fujiwara, Hikaru Shindo, Kaushalya Madhawa, Jun Jin Choong, Shuan Chen, Yuna Oikawa, Yiming Zhang, Gyubok Lee, Keisuke Ozawa
Triangular Resampling (TR) is a post‑training technique designed to reduce long‑horizon error accumulation in motion diffusion models by extending rollout‑based training to partially denoised states. TR uses a shared denoising threshold to replay multi‑step triangular denoising without gradient tracking, replacing states below the threshold with noise‑matched ground truth while retaining predictions above it. The method, applied to HumanML3D motion generation, achieves state‑of‑the‑art FID AUC, reducing it by 40.9% and the FID degradation slope by 55.3% compared to matched post‑training without replay.
By Kunhang Li, Yiyi Cai, Xiangyue Zhang, Fangyuan Tu, Yuhan Wu, Zhixiang Wang, Kaipeng Zhang, Haiyang Liu
Activation-Conditioned Self-Distillation (ACSD) is a new on‑policy self‑distillation method that uses a frozen copy of the base model to extract a steering vector by contrasting activations from self‑generated trajectories that reach verified correct answers with all other trajectories. The student learns from next‑token distributions on its own prefixes, without needing reference text or teacher parameter updates, and is used alone at inference. Across five models, ACSD achieves the highest mean accuracy on four mathematical benchmarks, with notable gains on DeepSeek‑R1‑0528‑Qwen3‑8B and LiveCodeBench v6 compared to the OPSD baseline.
By Zhexi Lu, Subhajit Chaudhury, Tejaswini Pedapati, Keerthiram Murugesan, Lei Yu
RetroGEF is a flow-based generative model designed for single-step retrosynthesis, which constructs possible reactants by adding atoms and altering bonds directly from a target molecule’s graph. Unlike traditional methods that rely on a fixed-size canvas or a prescribed edit order, RetroGEF models both connectivity changes and graph size variations within a single generative process, learning directly from product–reactant pairs. Experiments on standard retrosynthesis benchmarks show that RetroGEF achieves state‑of‑the‑art performance.
By Xiaozhuang Song, Xuemin Chen, Xinjian Zhao, Yaoyao Xu, Tianshu Yu
LEARN-TS is a method for multivariate time‑series anomaly detection that uses a frozen language model to generate two semantic representations: one that captures window‑specific observation context without exact numerical values, and another that provides a fixed, dataset‑agnostic normality prompt. These representations guide a patch‑masked reconstruction process, allowing the model to estimate normality discrepancy and produce timestamp‑level anomaly evidence. Experiments on four benchmarks show that LEARN‑TS achieves the best mean performance in 13 of 16 dataset‑metric comparisons, and ablation studies confirm the benefits of observation conditioning, joint normality alignment, and semantic references.
By Jahyeob Koo, Kio Yun, Byoungmo Koo, Jun-Geol Baek
QuanVI is a quantum-inspired algorithm for score-based variational inference that reformulates the optimization as a density-operator problem and uses a quantum tensor network parameterization with matrix product operators. By representing degenerate low-energy subspaces with maximally mixed states, it avoids the non-uniqueness of individual eigenvectors and compresses the density operator to prevent exponential parameter growth. Experiments demonstrate that QuanVI matches exact solutions in low dimensions and scales to high-dimensional synthetic and Bayesian posterior-approximation benchmarks, including challenging non‑Gaussian targets.
By Yuchen Cong, Zerui Tao, Chao Li, Zhe Sun, Qibin Zhao
The paper introduces a Physics‑Informed Method of Group Data Handling (PI‑GMDH) that builds functional representations of coupled physical fields progressively during the solution process, rather than fixing the structure beforehand. Using tensor‑product Chebyshev functions, the method is applied to the incompressible Navier‑Stokes equations in a two‑dimensional time‑dependent Taylor‑Green benchmark, achieving extremely low validation and test losses with a modest number of active functions. Comparative experiments with degree‑by‑degree, all‑terms PI‑GMDH, and reference PINN/KAN configurations demonstrate that selective progressive construction can yield a favorable balance of accuracy, model size, and computational time for this controlled synthetic benchmark.
By Mykhailo Minin