Large Reasoning Models (LRMs) have demonstrated remarkable capabilities by scaling up the length of Chain-of-Thought (CoT). However, excessively long reasoning traces pose substantial challenges for training cost and inference latency. While various CoT compression approaches have emerged to address this challenge, they face inherent trade-offs: token-level methods often disrupt syntactic and logical coherence, while step-level methods based on perplexity fail to reliably capture the logically critical reasoning steps because of the dilution of logical information. In this paper, we propose ASAP (Anchor-guided, SurprisAl-based Pruning), a novel coarse-to-fine framework for CoT compression. ASAP first performs anchor-guided pruning to preserve the core reasoning structure, which efficiently reduces the search space for subsequent processing. Leveraging the insight that logical branching choices are concentrated at the onset of reasoning steps, it then enables logic-aware pruning by selecting logically essential reasoning steps based on a novel first-token surprisal metric. Finally, ASAP distills these concise CoTs into target models so they can generate efficient reasoning traces at inference time. Experiments across math and code benchmarks show that ASAP preserves competitive accuracy while substantially reducing generated tokens, latency, and training cost, with additional validation on Qwen3 confirming its cross-model applicability.

Diffusion language models (DLMs) have recently emerged as an alternative modeling paradigm to autoregressive LMs, offering advantages such as parallel generation and bidirectional context modeling. Despite growing interest in their generative capabilities, the privacy risks of DLMs remain underexplored. We identify a phenomenon termed token-level memorization asymmetry through theoretical analysis of diffusion training dynamics. Building on this finding, we propose Q-Skew, a quantile-weighted skewness-based indicator for membership inference on finetuned DLMs. Experiments across multiple fine-tuning datasets and models show that our method outperforms existing baselines. Moreover, we show that Q-Skew can also facilitate other privacy violations, such as PII extraction. Our findings reveal a previously underexplored privacy attack surface and highlight the need for systematic privacy evaluation of DLMs.

Large Reasoning Models (LRMs) achieve remarkable performance by explicitly generating multi-step chains of thought, but this capability incurs substantial inference latency and computational cost. Collaborative inference offers a promising solution by selectively allocating work between lightweight and large models, yet a fundamental challenge remains: determining when a reasoning step requires the capacity of a large model or the efficiency of a small model. Existing routing strategies either rely on local token probabilities or post-hoc verification, introducing significant inference overhead. In this work, we propose a novel perspective on step-wise collaboration: the difficulty of a reasoning step can be inferred from its very first token. Inspired by the “Aha Moment” phenomenon in LRMs, we show that the entropy of the initial token serves as a strong predictor of step difficulty. Building on this insight, we introduce GlimpRouter, a training-free step-wise collaboration framework. GlimpRouter employs a lightweight model to generate only the first token of each reasoning step and routes the step to a larger model only when the initial token entropy exceeds a threshold. Experiments on multiple benchmarks demonstrate that our approach significantly reduces inference latency while preserving accuracy. For instance, GlimpRouter attains a substantial 10.7% improvement in accuracy while reducing inference latency by 25.9% compared to a standalone large model on AIME25. These results suggest a simple yet effective mechanism for reasoning: allocating computation based on a glimpse of thought rather than full-step evaluation.

Understanding and reasoning about entire soft-ware repositories is an essential capability for intelligent software engineering tools. While existing benchmarks such as CoSQA and CodeQA have advanced the field, they predominantly focus on small, self-contained code snippets. These setups fail to capture the complexity of real-world repositories, where effective understanding and reasoning often require navigating multiple files, understanding software architecture, and grounding answers in long-range code dependencies. In this paper, we present SWE-QA, a repository-level code question answering (QA) benchmark designed to facilitate research on automated QA systems in realistic code environments. SWE-QA involves 720 high-quality question-answer pairs spanning diverse categories, including intention understanding, cross-file reasoning, and multi-hop dependency analysis. To construct SWE-QA, we first crawled 77,100 GitHub issues from 12 popular repositories. Based on an analysis of naturally occurring developer questions extracted from these issues, we developed a two-level taxonomy of repository-level questions and constructed a set of seed questions for each category. For each category, we manually curated and validated questions and collected their corresponding answers. We evaluate six advanced LLMs on SWE-QA under various context augmentation strategies. Experimental results highlight the promise of LLMs.

Multi-agent systems powered by large language models have achieved strong performance on complex tasks, yet naive collaboration topologies often cause high communication costs and redundant context. Existing methods usually use a fixed communication graph and manage collaboration structure and shared memory in separate modules. Our log analysis of several representative systems shows that this separation leads to multiple copies of the same key facts in dialogue, memory and model inputs. We address this issue with EvoHyper, a framework based on an evolving hypergraph topology for multi-agent collaboration. In EvoHyper, a single hypergraph represents agents and shared memory, and each hyperedge serves as a collaboration unit that binds a group of agents to that shared memory. During execution a controller edits the hypergraph through a small set of predefined evolution operations, so collaboration units can spawn, update and merge as tasks unfold. Experiments on four benchmarks covering mathematical reasoning and code generation show that EvoHyper is (I) high-performing, achieving 3.2% to 7.8% accuracy gains over state-of-the-art methods, (II) efficient, reducing token consumption by up to 23.5%, and (III) adaptive, adjusting topology complexity according to task requirements.

Multimodal Large Language Models (MLLMs) have achieved remarkable performance in Visually Rich Document Understanding (VRDU) tasks, but their capabilities are mainly evaluated on pristine, well-structured document images. We consider document reconstruction from shredded fragments, a challenging VRDU setting that requires integrating visual pattern recognition with semantic reasoning under significant content discontinuities. To facilitate systematic evaluation of complex VRDU tasks, we introduce ShredBench, a benchmark supported by an automated generation pipeline that renders fragmented documents directly from Markdown. The proposed pipeline ensures evaluation validity by allowing the flexible integration of latest or unseen textual sources to prevent training data contamination. ShredBench assesses four scenarios (English, Chinese, Code, Table) with three fragmentation granularities (8, 12, 16 pieces). Empirical evaluations on state-of-the-art MLLMs reveal a significant performance gap: The method is effective on intact documents; however, once the document is shredded, restoration becomes a significant challenge, with NED dropping sharply as fragmentation increases. Our findings highlight that current MLLMs lack the fine-grained cross-modal reasoning required to bridge visual discontinuities, identifying a critical gap in robust VRDU research.

Long-horizon LLM agents often fail not because relevant information is absent, but because semantically relevant memories are epistemically invalid: stale observations, failed tool traces, or claims contradicted by newer verified evidence are retrieved and treated as support. We introduce HELM (Hierarchical Epistemic Learned Memory), a memory framework that makes this reliability problem explicit. HELM exposes memory through logged read/write/consolidate/prune operations and stores each item with provenance and epistemic state. Its three-tier Semantic-HNSW Nested Memory (SHNM) links episodic tool traces to consolidated recalls and thematic indices, enabling coarse-to-fine retrieval that can be audited back to concrete evidence spans. Epistemic governance filters and re-ranks recalls by status, freshness, consistency, recency, and provenance, while a lightweight learned controller decides when memory operations are worth their cost. Across five long-horizon benchmarks, HELM improves success over strong memory baselines on GAIA, WebArena, Mind2Web, StreamBench, and LifelongAgentBench while reducing effective context tokens and epistemic errors. New scaling, training-free, and governance-only ablations show that the main gains persist with a 32B backbone and that epistemic governance alone explains a large share of the reliability improvement.

Program verifiers play a central role in training coding agents, including selecting trajectories for supervised fine-tuning (SFT) and providing rewards for reinforcement learning (RL). Standard execution-based verification requires running unit tests inside per-repository environments such as Docker images, incurring substantial environment setup costs. We propose Dockerless, an environment-free agentic patch verifier that evaluates generated code patches without executing them. Rather than simply matching candidate patches to references, Dockerless judges patch correctness using evidence gathered through agentic repository exploration. On a verifier evaluation benchmark, Dockerless outperforms the strongest open-source verifier by 14.3 AUC points. Using Dockerless as both the SFT trajectory filter and the RL reward enables a fully environment-free post-training pipeline. The resulting model reaches 62.0%, 50.0%, and 35.2% resolve rate on SWE-bench Verified, Multilingual, and Pro, respectively. It surpasses the Qwen3.5-9B baseline by 2.4, 8.7, and 2.9 points, matching environment-based post-training.

Text-based role-playing models can imitate character styles, but often fail to capture scene atmosphere and evolving tension, which are crucial for immersive applications such as VR games and interactive narratives. We study video-grounded role-playing dialogue and introduce EBM-RL (Eye-Brain-Mouth Reinforcement Learning), a decoupled GRPO-based framework that separates observation ( lt;perception gt;), reasoning ( lt;think gt;), and utterance generation ( lt;answer gt;). This design mimics the human See-Think-Speak process, enabling the model to ground dialogue in visual perception before reasoning and response generation. To optimize this See-Think-Speak process, EBM-RL integrates complementary rewards for scene-text alignment, perceptual-cognitive utility, answer faithfulness, and format consistency. Extensive experiments show that EBM-RL substantially outperforms text-only role-playing baselines and comparable-scale VLM baselines, while remaining competitive with stronger larger-scale VLMs on visual grounding and situational persona metrics. Moreover, EBM-RL demonstrates strong zero-shot transfer to out-of-domain VideoQA benchmarks without additional fine-tuning. We also release an open-source dataset for video-grounded role-playing dialogue.

As Large Language Models (LLMs) evolve into autonomous agents for long-horizon tasks, managing unbounded interaction trajectories under fixed context budgets becomes a core systems challenge. Unlike standard long-context documents, agent trajectories are heterogeneous and interleave observations, reasoning traces, and tool executions, so compression must preserve temporal dependencies, actionable state, and structural fidelity. Yet existing methods remain fragmented, making it difficult to compare design choices and reason about their reliability implications. This survey introduces a unified taxonomy of agent context compression along three dimensions: compression target (what is compressed), compression mechanism (how it is transformed and retained), and control policy (who decides when compression is triggered). We further organize recurring failures in compressed execution into F1: Pre-compression Decision Error, F2: In-compression Information Loss, and F3: Post-compression Access Failure, and examine domain-specific trade-offs in software engineering, web navigation, and deep research. By unifying the design space, failure taxonomy, and evaluation perspective, this survey provides a foundation for building scalable and recoverable LLM agents.

The increasing size and complexity of large language models (LLMs) raise concerns about their ability to “cheat” on standard Question Answering (QA) benchmarks by memorizing task-specific data. This undermines the validity of benchmark evaluations, as they no longer reflect genuine model capabilities but instead the effects of data leakage. While existing methods detect such leakage, they fail to address the long-term challenge of mitigating it. In this paper, we introduce LastingBench, a novel approach to reinforce and safeguard existing benchmarks against knowledge leakage. Our method involves identifying leakage points through perturbation-based detection, followed by counterfactual rewriting to disrupt memorization while preserving the benchmark’s original evaluative intent. We demonstrate that our approach significantly reduces memorization effects in long-context QA benchmarks, providing a more accurate assessment of model reasoning and generalization abilities. Our experiments show that LastingBench not only uncovers substantial leakage in benchmarks like HotpotQA but also yields a more reliable evaluation of state-of-the-art models, ensuring that benchmarks remain effective and resilient over time.