Omni-modal large language models are increasingly evaluated on clean text–vision–audio inputs, where every channel is present, synchronized, and readily interpretable. Such scores are often taken as evidence of robust cross-modal fusion, but clean evaluation cannot tell whether success depends on stable cross-modal structure or on cues sufficient only in intact inputs. To address this gap, we define a modality fault line: a boundary at which model behavior becomes unstable when a modality remains present and human-interpretable, but its internal evidence structure is perturbed. We introduce SCEval (Structure-Corruption Evaluation) a diagnostic evaluation protocol that keeps the question, answer space, and modality channels fixed while applying controlled structural corruptions to text, vision, and audio individually and jointly. Built from 273 human-verified tri-modal examples from Social-IQ, OmniBench, and VALOR, SCEval evaluates 15 proprietary and open-source omni-modal systems. The results show that structural corruption lowers clean accuracy, text–vision damage forms the most stable shared fault line, and multi-modal degradation is non-additive rather than a simple function of the number of corrupted modalities. Clean omni-modal accuracy therefore does not establish that a model will remain reliable when cross-modal evidence becomes structurally unreliable.

Large language models (LLMs) excel at semantic understanding, yet their ability to reconstruct internal structure from scrambled inputs remains underexplored. Sentence-level restoration is difficult to evaluate automatically because scrambled sentences often admit multiple valid reorderings. We introduce OrderProbe, a deterministic benchmark for structural reconstruction using fixed four-character expressions in Chinese, Japanese, and Korean, which have a unique canonical order and thus support exact-match scoring. We further propose a diagnostic framework that evaluates models beyond recovery accuracy, including Semantic Accuracy, Logical Validity, Structural Consistency, Robustness, and Information Density. Experiments on twelve widely used LLMs show that structural reconstruction remains difficult even for frontier systems: zero-shot recovery frequently falls below 35%. We also observe a consistent gap between meaning-oriented generation and exact structural reconstruction, suggesting that structural robustness is not an automatic byproduct of semantic competence.

Video misinformation increasingly operates at the semantic and evidential level: authentic footage may be selectively edited, temporally reordered, spliced across sources, or augmented with AI-generated content to construct false narratives. Such evidence-dependent manipulations cannot be reliably verified from the input video alone, because the missing, reordered, replaced, or recontextualized evidence lies outside the video itself. We introduce EVID-Bench, a benchmark for search-grounded video misinformation detection, where a system must search the open web for related videos and identify what information is false through cross-video comparison. EVID-Bench comprises 222 videos spanning 9 manipulation types across 3 categories: AI generation, single-source editing, and multi-source editing. All samples are verified to be undetectable by frontier models through visual inspection alone. We evaluate nine frontier multimodal models using a retrieval-augmented verification baseline. The best system achieves only 61.43% point-level accuracy and 43.24% video-level accuracy, with AI-generated manipulations remaining especially difficult. Error analysis reveals recurring challenges: models fixate on irrelevant anchors, misattribute synthetic content to editorial splicing, and terminate search prematurely before fully explaining the manipulation.

Multimodal models often align language, vision, and audio in a single final-layer latent space, implicitly assuming that task-relevant evidence emerges at the same semantic depth across modalities. Using layer-wise CKA analysis, we observe that this assumption leads to semantic granularity mismatch: textual cues usually require deeper contextual abstraction, whereas visual and acoustic cues often provide discriminative perceptual evidence in shallow or middle layers. This mismatch can flatten fine-grained modality-private cues and reduce reliability under noisy, imbalanced, or missing inputs. To address this, we proposed Adaptive Hierarchical Representation Alliance (AHRA), a hierarchical shared–private expert framework. AHRA factorizes each modality into shared and private streams across semantic levels, regularizes them with shared alignment and private decorrelation, routes shared information through a cross-modal expert, and enhances task-relevant private tokens with modality-specific experts guided by a sparsity-controlled soft-gating mechanism (foreground exam). A hierarchical co-fusion module then performs intra-level expert coordination and inter-level semantic selection. Experiments on six benchmarks across image-text classification, multimodal intent recognition, and trimodal sentiment analysis show that AHRA consistently improves over strong baselines and remains robust under noisy and missing-modality settings.

The rapid proliferation of Multimodal Large Language Models (MLLMs) has ushered in the era of the “Agentic Economy,” where Mobile Agents autonomously execute high-stakes financial transactions. While these agents demonstrate impressive operational capabilities, their adversarial robustness remains a glaring blind spot. In this paper, we identify a systemic vulnerability termed Visual Dominance Hallucination (VDH), where imperceptible adversarial visual cues can act as a “super-stimulus,” overriding textual price evidence in our evaluated screenshot-based price-constrained settings and forcing the agent into irrational economic decisions. We propose PriceBlind, a stealthy, white-box adversarial attack framework for controlled screenshot-based evaluation. Unlike prior works that rely on conspicuous artifacts like pop-ups, PriceBlind exploits the modality gap in CLIP-based encoders via a novel Semantic-Decoupling Loss. Rather than literally making a luxury item “look cheap,” this regularizer weakens the consistency between high-price text and visual value cues by aligning the image embedding with a low-cost/value-associated anchor region while preserving pixel-level fidelity. On our main E-ShopBench benchmark with clear price constraints, screenshot-based white-box evaluation yields ASRs around 80% on the evaluated agents. Under the evaluated single-turn coordinate-selection protocol in a simplified layout-aware setting, our Ensemble-DI-FGSM strategy also yields non-trivial black-box transfer, with ASR roughly 35–41% across GPT-4o, Gemini-1.5-Pro, and Claude-3.5-Sonnet. In the same screenshot-based setting, standard robust encoders reduce ASR only partially, while a Verify-then-Act stack with robust encoders lowers ASR to below 10% at some clean-accuracy cost.

As Multimodal Large Language Models (MLLMs) become an indispensable assistant in human life, the unsafe content generated by MLLMs poses a danger to human behavior, perpetually overhanging human society like a sword of Damocles. To investigate and evaluate the safety impact of MLLMs responses on human behavior in daily life, we introduce SaLAD, a multimodal satety benchmark which contains 2,013 real-world image–text samples across 10 common categories, with a balanced design covering both unsafe scenarios and cases of oversensitivity. It emphasizes realistic risk exposure, authentic visual inputs, and fine-grained cross-modal reasoning, ensuring that safety risks cannot be inferred from text alone. We further propose a safety-warning-based evaluation framework that encourages models to provide clear and informative safety warnings, rather than generic refusals. Results on 18 MLLMs demonstrate that the top-performing models achieve a safe response rate of only 57.2% on unsafe queries. Morevoer, even popular safety alignment methods limit effectiveness of the models in our scenario, revealing the vulnerabilities of current MLLMs in identifying dangerous behaviors in daily life. Our dataset is available at https://github.com/xinyuelou/SaLAD.

Large language models are increasingly deployed as tool-augmented agents to acquire information beyond parametric knowledge. While recent work has improved long-horizon tool-use reasoning, most approaches focus on tasks with a single correct answer. In contrast, many real-world queries require discovering a comprehensive set of valid answers, a setting known as Multi-Answer QA. This setting raises two challenges: fine-grained credit assignment over long search trajectories and reward alignment for sustained exploration beyond easy high-frequency entities.We propose SPADER, a reinforcement learning framework for long-horizon tool use in Multi-Answer QA. SPADER includes Step-wise Peer Advantage (SPA), a critic-free step-level credit assignment mechanism that aligns parallel trajectories by decision step and estimates advantages from peer returns. It also includes a diversity-aware exploration reward that promotes long-tail entity discovery by upweighting rare findings and downweighting redundant ones.Experiments on QAMPARI, Mintaka, WebQSP, and QUEST show that SPADER generally improves recall and overall F1 over prompting-based agents, outcome-supervised RL methods, and recent step-level supervision approaches.Our code and model weights are available at https://github.com/KhanCold/spader.

Reinforcement learning from verifiable rewards (RLVR) drives chain-of-thought reasoning in large language models, yet its binary outcome reward cannot distinguish among correct trajectories. Existing dense reward alternatives, from surface heuristics to process reward models, either ignore the expert solutions already present in training corpora or require expensive offline annotation. We propose Gradient-Aligned Reward (GAR), which operates in the policy’s own gradient space: truncated backpropagation through the output projection layer extracts a compact gradient vector for each rollout, and cosine similarity with an expert-anchor gradient yields a dense, reasoning-aware reward with less than 9% wall-clock overhead. We prove that this cosine admits a multiplicative decomposition into prediction-error and activation-pattern factors, providing a concrete characterization of what the alignment signal measures. On Qwen3-4B and Qwen3-8B, GAR consistently improves over GRPO and other baselines on competition-level math benchmarks and transfers to GPQA Diamond and MMLU-Pro without domain-specific data. Code and data are available at https://github.com/LQgdwind/GAR .

The application of autonomous agents to scientific literature review is often constrained by the reliability of online search infrastructures and linear text generation. To address these limitations, we introduce SciLENS (Scientific Localized Evidence Navigation and Synthesis), an autonomous agent framework operating entirely upon a localized dual-tier data infrastructure for deterministic access to massive academic corpora. To mitigate online dependency, SciLENS utilizes a comprehensive offline toolbox featuring topological graph traversal and structural visualization. This empowers the agent to navigate deep citation networks and organize macro-level trends into structured visual representations, mitigating context exhaustion. To train this system without human annotation, we develop an automated multi-hop data synthesis pipeline validated by cross-model consensus, followed by rubric-based reinforcement learning to align the logical planning of the agent. Evaluations across standard tasks and newly constructed complex synthesis benchmarks demonstrate that SciLENS achieves robust factual precision, outperforming existing open-source baselines and achieving performance comparable to frontier proprietary models. Our source code and data are released at https://anonymous.4open.science/r/SciLENS-48DB .

Multi-agent language systems are often deployed under the assumption that collaboration improves reasoning quality, but this assumption breaks down once token cost, latency, and collaboration-induced errors enter the picture. We study protocol selection under a cost-aware utility objective, where the system decides on each input whether to run a collaborative protocol or fall back to solo reasoning. We propose Causal Advantage Routing for Language agents (CARL), a contextual-bandit framework that estimates the conditional collaboration advantage of each protocol over solo reasoning. CARL combines confounded execution logs with budgeted randomized probes, forms one-vs-solo doubly robust scores and shrunken DR-derived pseudo-outcomes, and at deployment uses a one-sided predictive lower bound for a future realized protocol-solo contrast, otherwise abstaining to solo execution. Across math, code, and multi-hop question-answering benchmarks with an open-weight model, CARL improves cost-aware utility and substantially reduces harmful collaboration over fixed protocols, compute-matched solo strategies, MAS routers, cascade methods, and causal routing baselines, indicating that effective multi-agent systems should estimate when collaboration is causally worth its cost rather than invoking it by default.

Criminal judgment prediction requires models to infer statutory articles, charges, and sentencing outcomes from case facts. Unlike standard classification tasks, it involves a structured reasoning process in which statutes should be matched with facts, charges should be justified by statutes, and sentencing outcomes should remain consistent with charges. Existing approaches optimize final labels, and while some have attempted to evaluate reasoning quality, their evaluations are indirect, often relying on LLM-generated rubrics that reflect model-internal preferences rather than the inherent logical structure of legal adjudication. We propose Juris Policy Optimization (JPO), a post-training framework for structured legal reasoning in Chinese criminal judgment prediction. JPO first uses teacher-generated rationales to supervise a standardized four-step reasoning process, and then applies reinforcement learning with a composite reward over legal prediction quality, reasoning structure completeness, and cross-step consistency. JPO further introduces token-level advantage reweighting and adaptive clipping for legally salient reasoning segments. Experiments on multiple open-source language models and three Chinese legal benchmarks show that JPO consistently improves both judgment prediction and reasoning quality over supervised fine-tuning and reinforcement learning baselines.

Shortcuts such as APIs and deep-links have emerged as efficient complements to flexible GUI operations, fostering a promising hybrid paradigm for MLLM-based mobile automation. However, systematic evaluation of GUI–shortcut hybrid agents remains largely underexplored. To bridge this gap, we introduce MAS-Bench, a benchmark that pioneers the evaluation of GUI-shortcut hybrid agents with a specific focus on the mobile domain. Beyond merely using predefined shortcuts, MAS-Bench assesses an agent’s capability to autonomously generate shortcuts by discovering and creating reusable, low-cost workflows. It features 139 complex tasks across 11 real-world applications, a knowledge base of 88 predefined shortcuts (APIs, deep-links, RPA scripts), and 9 evaluation metrics. Experiments demonstrate that hybrid agents achieve up to 68.3% success rate and 39% greater execution efficiency than GUI-only counterparts. Furthermore, our evaluation framework effectively reveals the quality gap between predefined and agent-generated shortcuts, validating its capability to assess shortcut generation methods. MAS-Bench addresses the lack of systematic benchmarks for GUI-shortcut hybrid mobile agents, providing a foundational platform for future advancements in creating more efficient and robust intelligent agents.

Continual learning (CL) for large language models (LLMs) aims to enable sequential knowledge acquisition without catastrophic forgetting. Memory replay methods are widely used for their practicality and effectiveness, but most rely on fixed, step-based heuristics that often misalign with the model’s actual learning progress, since identical training steps can result in varying degrees of parameter change. Motivated by recent findings that LLM forgetting mirrors the Ebbinghaus human forgetting curve, we propose FOREVER (FORgEtting curVe-inspired mEmory Replay), a novel CL framework that aligns replay schedules with a model-centric notion of time. FOREVER defines model time using the magnitude of optimizer updates, allowing forgetting curve-inspired replay intervals to align with the model’s internal evolution rather than raw training steps. Building on this approach, FOREVER incorporates a forgetting curve-based replay scheduler to determine when to replay and an intensity-aware regularization mechanism to adaptively control how to replay. Extensive experiments on three CL benchmarks and models ranging from 0.6B to 13B parameters demonstrate that FOREVER consistently mitigates catastrophic forgetting.

Recent Large Reasoning Models (LRMs) like DeepSeek-R1 have demonstrated remarkable success in complex reasoning tasks, exhibiting human-like patterns in exploring multiple alternative solutions. Upon closer inspection, however, we uncover a surprising phenomenon: The First is The Best, where alternative solutions are not merely suboptimal but potentially detrimental. This observation challenges widely accepted test-time scaling laws, leading us to hypothesize that errors within the reasoning path scale concurrently with test time. Through comprehensive empirical analysis, we characterize errors as a forest-structured Forest of Errors (FoE) and conclude that FoE makes the First the Best, which is underpinned by rigorous theoretical analysis. Leveraging these insights, we propose RED, a self-guided efficient reasoning framework comprising two components: I) Refining First, which suppresses FoE growth in the first solution; and II) Discarding Subs, which prunes subsequent FoE via dual-consistency. Extensive experiments across five benchmarks and six backbone models demonstrate that RED outperforms eight competitive baselines, achieving performance gains of up to 19.0% while reducing token consumption by 37.7%   70.4%. Moreover, comparative experiments on FoE metrics shed light on how RED achieves effectiveness.

The rapid advancement of large language models (LLMs) has not been matched by their evaluation in low-resource languages, especially Southeast Asian languages like Lao. To fill this gap, we introduce LaoBench, the first large-scale, high-quality, and multidimensional benchmark for assessing LLM language understanding and reasoning in Lao. LaoBench contains 17,000+ expert-curated samples across three dimensions: culturally grounded knowledge application, curriculum-aligned K12 education, and bilingual translation among Lao, Chinese, and English. It includes open-source and held-out subsets, where the held-out portion enables secure black-box evaluation via a controlled service to improve fairness and data security. We construct LaoBench with a hybrid pipeline that combines expert authoring with agent-assisted verification, ensuring linguistic accuracy, cultural relevance, and educational validity. We evaluate diverse state-of-the-art open-source and closed-source LLMs, and find that even strong multilingual models lag behind human experts, particularly in culturally grounded reasoning and translation fidelity. We hope LaoBench will catalyze research on Lao and other underrepresented Southeast Asian languages for more inclusive multilingual evaluation.

Large Reasoning Models (LRMs) have recently achieved remarkable success in complex reasoning tasks. However, closer scrutiny reveals persistent failure modes compromising performance and cost: I) Intra-step level, marked by calculation or derivation errors; II) Inter-step level, involving oscillation and stagnation; and III) Instance level, causing maladaptive over-thinking. Existing endeavors target isolated levels without unification, while their black-box nature and reliance on RL hinder explainability and controllability. To bridge these gaps, we conduct an in-depth white-box analysis, identifying key neurons (Mixture of Neurons, MoN) and their fluctuation patterns associated with distinct failures. Building upon these insights, we propose NeuReasoner, an explainable, controllable, and unified reasoning framework driven by MoN. Technically, NeuReasoner integrates lightweight MLPs for failure detection with a special token-triggered self-correction mechanism learned via SFT. During inference, special tokens are inserted upon failure detection to actuate controllable remedial behaviors. Extensive evaluations across six benchmarks, six backbone models (8B 70B) against nine competitive baselines, demonstrate that NeuReasoner achieves performance gains of up to 27.0% while reducing token consumption by 19.6%   63.3%.

Multimodal Large Language Models (MLLMs) have demonstrated significant advances across numerous vision-language tasks. Due to their strong performance in image-text alignment, MLLMs can effectively understand image-text pairs with clear meanings. However, effectively resolving the inherent ambiguities in natural language and visual contexts remains challenging. Existing multimodal benchmarks typically overlook linguistic and visual ambiguities, relying mainly on unimodal context for disambiguation and thus failing to exploit the mutual clarification potential between modalities. To bridge this gap, we introduce MUCAR, a novel and challenging benchmark designed explicitly for evaluating multimodal ambiguity resolution across multilingual and cross-modal scenarios. MUCAR includes: (1) a multilingual dataset where ambiguous textual expressions are uniquely resolved by corresponding visual contexts, and (2) a dual-ambiguity dataset that systematically pairs ambiguous images with ambiguous textual contexts, with each combination carefully constructed to yield a single, clear interpretation through mutual disambiguation. Extensive evaluations involving 19 state-of-the-art multimodal models—encompassing both open-source and proprietary architectures—reveal substantial gaps compared to human-level performance, highlighting the need for future research into more sophisticated cross-modal ambiguity comprehension methods, further pushing the boundaries of multimodal reasoning.

Multimodal large language models (MLLMs) have demonstrated promising results in a variety of tasks that combine vision and language. As these models become more integral to research and applications, conducting comprehensive evaluations of their capabilities has grown increasingly important. However, most existing benchmarks fail to consider that, in certain situations, images need to be interpreted within a broader context. In this work, we introduce a new benchmark, named as CODIS, designed to assess the ability of models to use context provided in free-form text to enhance visual comprehension. Our findings indicate that MLLMs consistently fall short of human performance on this benchmark. Further analysis confirms that these models struggle to effectively extract and utilize contextual information to improve their understanding of images. This underscores the pressing need to enhance the ability of MLLMs to comprehend visuals in a context-dependent manner.