Agent 进展

Agentic retrieval-augmented generation (RAG) systems can fail before evidence-conditioned reasoning is tested: an agent may retrieve candidate snippets but finalize without inspecting them. We study this failure mode as a procedural property of the agent trajectory, decomposing wrong answers into pre-evidence discipline failures and post-gold-read failures using saved tool-call traces, retrieved evidence, read passages, and final answers. Across 12,000 paired trajectories on HotpotQA, 2WikiMultiHopQA, and MuSiQue, the two failure types are largely non-redundant: the both-trigger rate is in [11.2%, 13.1%] across regex and spaCy entity extractors. We then evaluate Read-Gate, a minimal runtime invariant requiring an agent to read after search and before finalization. Forced reading improves LLM-Acc by 14.9-19.9 points on trajectories that would otherwise skip reading and by 3.2-9.4 points on full minimal-reasoning cells. Additional diagnostics show that larger hidden thinking budgets do not necessarily increase evidence inspection. Together, these results indicate that evidence-gathering should be evaluated as a trajectory-level control problem, separately from answer-side reasoning.

LLM-agent evaluations often produce task outcomes long before the full benchmark run is complete. A partial score is tempting to report, but it does not show whether the observed tasks support the same conclusion as the completed evaluation. Early tasks can omit important parts of a benchmark, running cheaper tasks first can distort the observed sample, and a rule that decides only easy pairs can appear accurate while leaving many comparisons unresolved. We introduce ParEvalLayer, a decision layer that reads paired outcomes for two agent systems and a comparison policy chosen in advance. For each partial run, it records whether the tested agent system is better by the required amount, is not better by that amount, needs more evidence, or should abstain. We evaluate ParEvalLayer by replaying completed public benchmark data as if each evaluation had stopped earlier. At each point, ParEvalLayer applies the policy using only the outcomes observed so far; if it reaches one of the two comparison judgments, we check whether that judgment matches the completed data for the same system pair. With the main comparison rule, three of the public benchmarks reach the same decision as the completed evaluation after observing only 15% to 25% of task outcomes. Other benchmarks require more task outcomes. This variation shows why a partial score alone is not enough: reports should also state the decision rule and how many comparisons remain without a decision.

LLM agents fail mid-episode — they loop, cascade tool errors, drift off goal, fabricate results, or silently absorb corrupted content — and the standard remedy, judging every step with a second LLM, costs more than the agent itself. We ask how much detection is achievable from observable step telemetry alone, using monitors costing microseconds per step and trained only on healthy runs. On 2,823 committed agent episodes across three frameworks, three local models (qwen2.5 7b/3b, llama3.1 8b) and a commercial API (gemini-2.5-flash), a one-class echo-state-network ensemble with CUSUM alarms detects 0.71 of failures at a 5% false-alarm budget (AUROC 0.872). Its advantage over a memoryless baseline is a monotone function of post-onset horizon (+0.09 at <=3 steps, +0.40 at >=9), predicting its own failure region out of sample on AFTraj-2K. Ranking transfers with no retraining to two corpora from other groups (AFTraj-2K 0.745, ATBench 0.779). Monitors carry two burdens: a per-deployment healthy null (they do not transfer — AUROC 0.527 cold against 0.885 recalibrated) and a residual false-alarm rate. We add a layer carrying neither: deterministic verification, which recomputes a run's stated total from the tool results it actually received and confirms every required call was made. Head-to-head it catches 60% of failures (96% with the coverage check) at 0 of 63 false positives against the monitor's 54% at 17%, transfers unchanged to llama3.1:8b (110 of 110 at 0 of 10), and trips on 0 of 1825 healthy episodes. Detection is then closed into repair: each flagged run is rolled back and re-run live, recovering 45% of failures against a 16% resampling control (p=0.0005) and lifting task success from 52% to 73% for about one extra model call per run. The system runs at ~200 microseconds per step, three orders of magnitude below a judge call. Code, traces and results are released.

Persistent memory allows (self-evolving) LLM agents to adapt across tasks by consolidating heterogeneous interaction histories into reusable facts, preferences, observations, and rules. Yet consolidation also imposes an implicit authorization boundary: it determines whether stored information may later be consumed as a user fact, an attested observation, or a standing instruction. We identify authority collapse, in which consolidation preserves a claim while erasing the source constraints governing its authorized use, causing the stored memory to imply greater authority than its source permits. We introduce AuthMem-Bench, a controlled paired benchmark that holds the focal claim and downstream task fixed while varying only source authority. It evaluates write-time collapse, downstream authorization errors, and automatic authority preservation. Across seven consolidators based on widely used agent-memory systems and seven LLM backbones, we observe authority collapse in 48 of 49 evaluated configurations. In a controlled action-grounded evaluation, collapsed memories without authority metadata yield a mean unauthorized-action rate of 50.3%. In an end-to-end evaluation, automatically predicted and persisted authority labels reduce the observed unauthorized-action rate from 16.9% to 0.0%, while benign task success remains essentially unchanged. These findings show that memory-driven adaptation must preserve not only what was learned, but also the authority under which it may be reused.

Large language models (LLMs) are increasingly extended into deep search agents that solve complex questions through multi-step interaction with external search and browsing tools. However, existing agents often incur substantial computational and interaction costs, generating lengthy trajectories that contain redundant queries, inefficient exploration, and irrelevant observations. Existing efficiency-oriented methods usually encourage agents to use tools less frequently, but treating all tool interactions uniformly may also suppress steps that gather necessary evidence. In this paper, we propose CRISP, a framework for training efficient deep search agents through critical step perception. Unlike prior efficiency methods that uniformly penalize tool use, CRISP distinguishes interactions that gather necessary evidence from redundant ones and shapes the training reward to preserve the former while pruning the latter, improving efficiency without sacrificing the evidence needed for correct answers. Specifically, CRISP first constructs critical-step labels with Backward Evidence Induction: starting from the final answer, a strong model traverses a completed search trajectory backward and judges whether each tool-interaction step provides or preserves evidence for the final answer. We then distill these step-wise judgments into a smaller critical-step recognizer, enabling full-trajectory analysis in a single pass. During policy optimization, an efficiency-aware reward is applied only to successful rollouts. Experiments on BrowseComp and HLE-Verified show that CRISP maintains competitive final-answer accuracy while reducing average interaction turns by 15.1% and 33.2%, respectively, demonstrating substantial improvements in interaction efficiency.

Harness design plays a critical role in agent performance by shaping how large language models (LLMs) perceive, reason over, and act within executable environments. Recent work has proposed automatic harness evolution, which iteratively improves the harness from agent--environment interactions. However, existing methods often overfit to the evolution tasks, rely exclusively on trajectory-derived signals, and optimize harness components jointly, causing interference across components. We propose HarnessCompass, a novel automatic harness evolution framework built around constrained evolution, proactive feedback, and component-wise optimization. HarnessCompass first enforces global constraints on evolution, restricting modifications to task-agnostic harness changes that generalize beyond the evolution tasks. It then augments trajectory-derived evidence with proactive first-person feedback from the agent about harness usage, yielding richer signals for evolution. Finally, it decouples the optimization of different harness components before consolidating them into a unified harness, reducing cross-component interference while preserving component synergy. On SWE-bench Verified with GPT-5.4, HarnessCompass improves Pass@1 from 54% to 66% in only 5 evolution iterations, outperforming AHE in both effectiveness and evolution efficiency. In addition, the evolved harness transfers effectively to held-out tasks and other models, demonstrating substantially stronger generalization than prior automatic harness evolution methods.

Tool-using large language model (LLM) agents produce long, multi-turn trajectories, making gradient-based post-training memory-intensive. Evolution strategies (ES) enable memory-efficient full-parameter post-training without backpropagation and can eventually match the performance of gradient-based reinforcement learning (RL). However, resource-constrained settings typically offer only a few GPUs, so the high GPU-hour requirements of ES translate into prohibitively long training times. To address this, we introduce Cooperative Parameter-subspace Evolution Strategy (CoPES), a cooperative coevolutionary method that decomposes the full parameter space into lower-dimensional subspaces and searches over them cooperatively to improve optimization efficiency. We post-train a Qwen3.5-4B tool-using agent for the math task and evaluate it on five benchmarks of varying difficulty. Under the GPU-hour budget of full-parameter GRPO's best validation checkpoint, CoPES recovers 92% of GRPO's validation-accuracy gain, versus 67% for standard ES, while its theoretical GPU memory requirement is less than one-eighth that of full-parameter GRPO. It consistently outperforms standard ES and LoRA-based GRPO on all evaluated pass@k metrics across the five benchmarks. Additional experiments further show the advantage of CoPES on the question-answering task. These results demonstrate an improved trade-off between memory requirements and training time for agentic LLM post-training under resource constraints. The code is open-sourced in https://github.com/MetaronWang/CoPES

As memories accumulate across tasks and sessions, the performance of long-term LLM agents depends jointly on query-specific retrieval and continual memory refinement. However, existing methods typically optimize either memory access, through iterative query refinement or adaptive retrieval policies, or memory evolution such as structural update. This separation overlooks a fundamental feedback loop: retrieval determines which memories receive usage signals, while updated memory bank reshape future retrieval. We propose CoEvo-Mem, a closed-loop framework for co-evolving the retrieval policy and memory bank. For each query, a frozen LLM generates route-specific query rewrites and a routing prior, which a lightweight residual router corrects online. The retrieved context serves as the coupling interface between the two learning processes: task outcomes assign credit to routing decisions, while trajectory-conditioned feedback updates memory values and graph relations. These updates alter how memories are ranked and selected for subsequent queries, thereby closing the feedback loop. To mitigate coupling induced non-stationarity, CoEvo-Mem alternates between updating the router with the memory bank fixed and evolving the memory bank with the retrieval policy fixed. Across seven diverse benchmarks, CoEvo-Mem achieves state-of-the-art performance, demonstrating the importance of retrieval-memory coevolution.

Multimodal large language models (MLLMs) have advanced visual understanding and reasoning, yet their static parametric knowledge limits their ability to address knowledge-intensive and dynamically evolving open-world problems. To move beyond this limitation, multimodal deep search has emerged as a key direction for open-world information access, evolving from single-turn factual retrieval toward long-horizon, multi-turn search guided by visual evidence. However, existing methods typically confine vision to the input or answer stage, overlooking its role in intermediate reasoning, and lack designs tailored to long-horizon interaction. Consequently, visual evidence rarely drives continued retrieval, constraining both interaction depth and reasoning span. To address these limitations, we propose DeepVoyager-VL, a long-horizon multimodal deep-search framework for vision-in-the-loop search. Specifically, we construct a multimodal event graph to drive data synthesis, yielding problems with intermediate visual dependencies and long reasoning chains. We then design an agent framework for active visual acquisition and on-demand image loading. Finally, we fine-tune models on the synthesized data without reinforcement learning. Extensive experiments across ten multimodal search benchmarks demonstrate the effectiveness of our method.

Large language model (LLM) agents must retain and use cross-step information to act coherently in long-horizon tasks. Existing methods improve memory accessibility, yet action-relevant information may still fail to guide the current decision because it is poorly formed, organized, prioritized, or presented. We call this post-access failure the Memory-Action Gap. We propose MemArbiter, a function-aware memory arbitration framework that addresses the memory-management-induced component of this gap. MemArbiter decomposes interaction histories into atomic items, organizes them into five functional Memory Banks, and combines bank-level demand, item-level relevance, focal-ambient representations, and a temporal presentation gate to dynamically control memory salience. We evaluate MemArbiter on ALFWorld against Flat Retrieval and Flat Recency under unified per-step memory budgets. With an open-weight action-generation model, MemArbiter achieves success rates of 82.8% and 92.5% under 500- and 750-token budgets, outperforming the strongest baseline by 20.9 and 25.4 percentage points, respectively. It also improves post-failure recovery and reduces failed-action repetition and state-action recurrence. These results show that function-aware memory arbitration enables accessible information to guide actions more effectively.

Tool-using large language model agents frequently replan, retry failed operations, delegate tasks, and resume after crashes. These behaviors can cause one user authorization to be requested and executed multiple times under freshly issued token identifiers, even when each individual token is single-use. We call this failure semantic replay: exceeding the execution budget of a token-independent authorization instance rather than merely reusing an old token identifier. We show that identifier-local token consumption cannot prevent fresh reissuance unless the issuer retains monotonic durable state over the authorized action, confirmation event, and remaining execution budget. We introduce CapLease, an authorization-consumption layer that follows proposal- and authority-level defenses, binds an authenticated user confirmation to a canonical action, and enforces transactional Issue-Prepare-Commit transitions. Across LLM-agent replanning, retry, delegation, concurrency, confirmation-replay, and crash-recovery scenarios, identifier-local tokens permit fresh semantic reissuance, whereas CapLease and an equally stateful Server Ledger prevent duplicate admission and, with an idempotent sink, duplicate external effects. Our results identify durable authorization state, rather than token representation alone, as the systems requirement for replay-resistant agent execution.

Training LLM-based search agents requires high-quality search data: tasks that demand genuine multi-hop retrieval and trajectories that use search tools effectively. Existing pipelines often depend on human-written tasks, expert demonstrations, or stronger teacher models. We present SearchMaster, a self-play framework that trains a single LLM from search tasks it generates, solves, and verifies in a local search environment. The key challenge is that self-generated tasks and rollouts can yield misleading signals: pseudo multi-hop questions, success-rate difficulty estimates that ignore search depth, and rollouts with excessive opening but little targeted evidence acquisition. SearchMaster addresses these failure modes with three controls. An Evidence-Chain Generator (ECG) grounds task generation in explicit cross-document evidence chains to reduce pseudo multi-hop questions. A Search-Depth Reward (SDR) scores task difficulty by the search depth of successful rollouts rather than success rate alone, keeping retained tasks search-intensive. An Over-Opening Penalty (OOP) regulates tool use by discouraging excessive document opening, avoiding long but shallow browsing. Verified Proposer and Solver rollouts are then jointly optimized with GRPO. Across six deep-search benchmarks, SearchMaster improves a Qwen3.5-9B backbone from 38.19% to 51.52% average accuracy, with a 30.1-point gain on BrowseComp-Plus. These results show that grounded and regulated self-play can provide effective search-agent training data without human-labeled QA pairs or expert demonstrations. The code is available at https://github.com/WentaoTan/SearchMaster.

Agentic multimodal reasoning extends passive image understanding by allowing VLMs to actively acquire and refine visual evidence through visual tool interactions. Effective visual tool use requires three capabilities: grounding tool calls in visual context, composing tools across multiple steps, and adapting reasoning to tool-returned observations. However, existing approaches largely focus on grounding within fixed tool spaces and rigid invocation patterns, leaving composition and adaptation insufficiently addressed. We present VC-Tooler, which learns visual tool use as a compositional and adaptive capability. To this end, we first build a trajectory bank through a hierarchical synthesis pipeline covering three capability levels: single-tool grounding, multi-tool composition, and diverse tool contexts and interfaces. We then train the model in two stages: a supervised cold start that establishes these capabilities, followed by reinforcement learning that encourages accurate, efficient, and context-aware visual tool use. VC-Tooler achieves state-of-the-art performance among open-source models on both general-purpose and agentic benchmarks, including $95.8%$ on V* and $35.3%$ on VTC-Bench, and shows promising transfer under richer tool settings at inference time. Project page: https://w1zheng.github.io/VC-Tooler

The rapid development of photonic integrated circuits (PICs) is shifting the design flow from traditional graphical user interface (GUI)-based methods to script-based methods for higher flexibility, portability, and maintainability. However, script-based design introduces new challenges, requiring designers to possess additional proficiency in tool application programming interfaces (APIs) and programming. It also demands greater effort and time because it is inherently less intuitive and more complex than GUI-based methods. As PICs grow in scale and complexity, the productivity gap between design needs and manual scripting capabilities continues to widen. To address this gap, we introduce PICopilot, the first large language model (LLM)-based agentic framework that assists in PIC design via automated design script generation from natural language instructions. PICopilot leverages a multi-agent architecture with a feedback mechanism and a specifically designed retrieval-augmented generation (RAG) pipeline, achieving a high success rate and reliability. Experimental results on a benchmark of diverse PIC scripting tasks demonstrate that PICopilot successfully completes all 48 tasks and outperforms other LLM-based approaches without incurring substantial extra latency or cost, even solving 21 more tasks than the advanced GPT-5 model with a general RAG pipeline.

Native computer use offers a general interface for agents to operate almost any software available to people, but requires long-horizon state tracking, large-scale interactive experience, and learning from sparse yet verifiable outcomes. We introduce Qwen-CUA, a native computer-use agent with a 397B-A17B Qwen mixture-of-experts backbone. It observes only screenshots and acts through keyboard and mouse events, without DOM trees, accessibility metadata, or task-specific APIs. Its scaffold maintains up to 20 active screenshots and folds older visual history in fixed-size blocks to retain recent evidence while preserving reusable prompt prefixes. For training, we build a cloud rollout fleet with access to nearly 100,000 vCPUs and tens of thousands of concurrent environments, construct approximately 40,000 verifiable tasks, and collect personalized long-horizon workflows across everyday and professional software. We optimize complete trajectories with verifiable rewards and trajectory slicing, while iterative training runs refresh supervised data and recalibrate reinforcement-learning tasks. Across eight benchmarks, Qwen-CUA outperforms Qwen3.7 and remains competitive with leading proprietary systems, reaching 86.2 on OSWorld-Verified and 18.5/48.4 binary/partial completion on OSWorld 2.0. Scaling the same recipe to a model with over one trillion parameters yields Qwen-CUA-Max, improving these scores to 87.6 and 21.2/53.3. Qwen-CUA also reduces RedTeamCUA attack success from 36.6 to 16.4 relative to Qwen3.7. Efficiency analyses, a browser deployment, and Bash-augmented experiments further characterize practical behavior. These results establish native computer use as a broadly capable agent foundation and highlight scalable verifiable interaction and hybrid tool use as key directions.

Computer-use agents (CUAs), which empower large language models to autonomously operate operating systems and the web, are increasingly vulnerable to indirect prompt injection attacks. A widely adopted defense is the human-in-the-loop paradigm, in which the agent pauses for explicit user confirmation before executing sensitive operations. While effective against conspicuously high-harm attacks, this defense offers little protection against what we term Invisible Ink Threats: low-harm injected goals, such as starring a repository or installing a package, that are behaviorally indistinguishable from legitimate task execution and thus evade both model safety mechanisms and human oversight. To systematically investigate this blind spot, we present II-Bench, a collection of seemingly harmless adversarial tasks. II-Bench comprises 444 examples targeting confidentiality and integrity attacks across three platforms, spanning three attack categories: page navigation and interaction, sensitive information exfiltration, and code download and execution. Each category is instantiated in both natural language and code forms under two levels of instruction specificity. Furthermore, we construct HITLCUA, a comprehensive adversarial testing framework that integrates a real virtual machine operating system environment with isolated Docker-based web platforms, and simulates human participation by allowing CUAs to consult an API-simulated user before proceeding with suspicious operations. Extensive evaluations of leading CUAs reveal that low-harm injections frequently bypass both agent defenses and simulated user review, exposing severe and previously underexplored security risks in current CUAs.

Learning-based memory systems for self-evolving LLM agents face two tightly coupled challenges. First, trajectory-indexed utilities grow with the interaction history, thereby dispersing limited feedback over an ever-expanding state space. Second, because trajectory-level rewards are jointly assigned to co-retrieved memories, irrelevant experiences may receive misleading utility updates and consequently enter the memory-reward trap. To address these challenges, we introduce Reduced-Order Memory Reinforcement Learning (RoMeRL), which represents the growing trajectory-indexed utility space using a fixed-dimensional per-task memory state factorized by outcome polarity and memory dynamics. RoMeRL incorporates new experiences through a fixed set of semantic coordinates whose contents are updated or replaced over time, thereby concentrating feedback over a bounded utility support. Theoretically, we show that this reduced-order parameterization increases the average feedback received by each utility coordinate and characterize the steady-state occupancy of erroneous coordinates under a generic coordinate-transition model. Empirically, across ALFWorld and LifelongAgentBench, RoMeRL improves task performance, reduces the Cold-Q ratio by 80.0%, increases feedback density by approximately 6.0 times, reduces the maintained memory size by 84.4%, and cuts LLM calls by 21.1%. These results show that reduced-order utility states support efficient self-evolving agent memory while limiting persistent reward contamination. Code is available at: https://github.com/YOUNG-fnxm/RoMeRL

Traditional software cost estimation models, such as COCOMO II, Function Points, and Story Points, assume that development effort is primarily driven by human labor in design, coding, and testing. Agentic software engineering, where autonomous AI agents perform substantial implementation work and humans focus on planning, specification, and validation, challenges this assumption. New cost dimensions arise: large language model (LLM) token consumption across agent actions, Human-in-the-Loop (HITL) oversight effort, and infrastructure costs for agent orchestration and tooling. These costs are nondeterministic: identical tasks may consume different tokens, follow divergent reasoning paths, and require varying human correction, phenomena absent in traditional development. A new framework is needed to bridge standard sizing metrics with this cost structure. This paper proposes ACEM (Agentic Cost Estimation Model), which decomposes total agentic development cost into three additive dimensions: LLM, HITL, and infrastructure cost. ACEM introduces three constructs for agentic dynamics: the Revision Factor (RF), modeling token overhead from output rejection and retries; the Context Factor (CF), capturing rising token consumption as context accumulates; and the HITL Intensity Score (HIS), a four-level oversight classification scheme. It further maps Use Case Points, Story Points, and Function Points to estimated token consumption, enabling organizations to reuse existing project-scoping data for agentic cost forecasting. ACEM is presented as a fully specified model structure and calibration methodology, with constants left symbolic pending empirical grounding. As an early-stage proposal, it invites the research community to calibrate, test, and extend the model through real project data.

Scientific poster construction compresses a long multimodal paper into a readable, editable canvas. Existing systems hide request-level failures by scoring only completed outputs; direct image generation is not element-editable, while coding-agent workflows are costly. PosterMELD is a template-conditioned multi-agent pipeline: capacity-aware slots guide writing before rendering, and deterministic gates plus vision-language model (VLM) review route failures to bounded repair. Each accepted request exports editable PowerPoint (PPTX) and Portable Network Graphics (PNG) artifacts; explicit design controls yield same-paper variants. Across 621 papers, Print-Ready Rate (PRR) counts requests passing geometric, readability, asset-integrity, and obvious-factual-error checks, with native editability reported separately. A frozen VLM assigns conditional Craftsmanship-Harmony-Expressiveness (CHE) scores to print-ready outputs. PosterMELD attains 81.3% PRR, 3.4 times P2P's rate and 5.2 times PosterGen's, and the highest conditional CHE among generated methods with multiple print-ready outputs. Native editability and explicit design controls are retained at a mean cost of USD 0.38 per request, 3.5% of Codex+Skill's. Code and resources are available at https://github.com/Shannon4Science/PosterMELD.

Learning to adapt strategies through interaction is a key step toward more general and autonomous LLM agents. Existing approaches typically achieve behavioral adaptation by revising skill libraries. However, in multi-agent environments, opponents may simultaneously update their strategies, causing the environment itself to evolve continuously. Applying skill-revision methods designed for static environments in such settings therefore amounts to updating against an obsolete reference. To address this challenge, we introduce OASE (Opponent-Aware Selective Evolution), which identifies and adopts genuinely beneficial skill revisions in dynamic multi-agent environments. Specifically, OASE conducts paired comparisons between a candidate skill and the incumbent under identical conditions anchored by historical snapshots of opponent strategies, and adopts the candidate only when its estimated payoff gain exceeds an acceptance threshold. We evaluate OASE in two decision-making scenarios: first-price auctions and private-cost Cournot competition. Experimental results show that, compared with a Reflexion-style baseline, OASE achieves a lower final equilibrium distance in both environments while accepting substantially fewer skill revisions, thereby suppressing strategy changes that lack sufficient payoff support. OASE therefore replaces blind updating with evidence-anchored selection, allowing agents to adapt stably and efficiently even as opponents continuously evolve.

Large language model (LLM) agents increasingly undertake long-horizon tasks that require sustained reasoning, tool use, and revision across many interdependent steps. However, existing agent harnesses maintain task execution, task state, and completion assessment within a growing context, making the state difficult to track and allowing incorrect self-assessments to propagate into later decisions. We reformulate long-horizon execution as a task-state management problem and propose LongHorizon-Harness, which maintains the task state explicitly outside execution and updates it only with facts independently verified from the environment. Its Manage-Execute-Audit(MEA) loop uses a manager to maintain the task state and determine the next subtask, a fresh-context executor to perform it, and a read-only auditor to verify the resulting environment state before the next round. A lightweight AgentAdapter supports interchangeable model and harness backends without modifying their native agent loops. LongHorizon-Harness improves Qwen~3.7-Plus from 51.8% to 80.7% on WeaveBench, from 69.7% to 77.2% on Terminal-Bench~2.1, and from 2.8% to 8.3% on OSWorld~2.0. It also raises Claude Opus~4.7 from 20.0% to 34.3% on an OSWorld2.0 subset, demonstrating consistent gains across models, harnesses, and interaction domains.

Complex acoustic problems may require models to perform acoustic operations, interact with external tools and reason over the resulting textual or processed-audio observations rather than answer directly from a fixed audio input. We study such problems as tool-interactive audio reasoning and develop SpeechAgent-R, an audio agent that coordinates its intrinsic multimodal understanding with external skills and tools. To support this capability, we construct HIU-Corpus, comprising 65,492 interaction trajectories and 507.6 hours of audio across 24 tasks, 8 skills and 9 tools. SpeechAgent-R first learns structured interaction behaviors through trajectory-based supervised fine-tuning and then improves its decisions through multi-turn reinforcement learning. We further introduce HIU-Bench to jointly evaluate task performance, interaction quality and generalization to diverse task settings. It contains 1,395 samples across 56 tasks, including in-distribution (ID) and out-of-distribution (OOD) splits with substantial shifts in tool usage and workflow composition. SpeechAgent-R achieves 84.17 on ID tasks and 70.94 on OOD tasks, improving over the base model under the same agent harness by 15.40 and 14.23 points. These results demonstrate that learning skill and tool coordination improves audio agents' ability to handle diverse task settings and adaptive tool interactions.

We study what happens when a single general-purpose large language model acts as the sole researcher on a long-horizon neural architecture design problem. The agent receives a scientific question, an initial hypothesis and motivation, a compute budget, and research affordances (source and experiment management, experiment tracking, literature access, and persistent memory), then autonomously proposes, implements, evaluates, and records experiments over an extended period. The study comprises three phases, separated by human-declared transitions, that progressively expand the agent's tool surface or problem scale. Across approximately 100 sequential experiments, the agent improves a non-standard Vision Transformer from a weak baseline to a stronger, efficient model on small benchmarks and a usable but sub-SOTA model on ImageNet-1K, while producing a dense behavioural trace. We report four findings.(i)Productivity exhibits a clear phase structure: rapid early gains, a multi-dozen-hypothesis saturation wall, and recovery, with recovery triggered by expanding the action surface rather than changing the underlying model.(ii)A single early hypothesis contributes more to accuracy gain, with later improvements long-tailed.(iii)The preference for greedy, incremental hypotheses is largely workflow-induced: a commit-or-discard evaluation rule is isomorphic to greedy hill-climbing; the remainder reflects risk aversion after bold failures and anchoring on familiar literature. (iv)The agent independently rediscovers established results and, in the unfamiliar regime of pure channel attention, overturns a standard design choice. We conclude that workflow design was at least as influential as agent capability in this study and propose diversified search, budgeted moonshot hypotheses, explicit forks, and regime-aware re-validation as testable directions for future autonomous research.

Large Language Models have enabled increasingly capable autonomous agents, yet personalization remains critical for making such agents practically useful. Recent benchmarks have begun evaluating personalization in agents, but they largely rely on static preference snapshots, fixed interaction logs, or question answering over predefined user profiles. Such designs fail to capture the complexity of evolving user preferences and neglect preference-conditioned task execution-a discrepancy we term as the knowledge-to-action gap. To address this challenge, we introduce IBA-Bench, a benchmark for implicit behavioral alignment constructed from longitudinal interaction histories that contain noise, implicit cues, and temporal inconsistencies. Unlike prior work, IBA-Bench evaluates whether an agent can execute tasks while satisfying implicit user constraints inferred from historical interactions. We further propose IBA-Agent, an agent framework that reconciles conflicting priorities through broad retrieval and trajectory-level alignment. Experiment results on IBA-Bench show that effective personalization remains a significant challenge for state-of-the-art LLM agents, and the proposed IBA-Agent substantially improves behavioral alignment in complex scenarios across nine application domains.

Multi-agent reasoning (MAR) improves reasoning reliability through iterative solution exchange and refinement. Existing adaptive MAR methods typically learn routing decisions from query-level labels or trajectory-level returns, but such coarse supervision cannot accurately estimate the state-conditioned utility of individual operators in multi-step collaboration. We propose TreeCredit, a shared-prefix credit assignment framework for efficient adaptive MAR. Its core insight is to estimate operator utility through state-matched downstream comparisons, rather than directly attributing trajectory-level outcomes to preceding decisions. TreeCredit constructs shared-prefix collaboration trees by expanding candidate operators from the same intermediate state and assigns each state--operator pair a correctness-prioritized suffix credit based on the terminal correctness and cumulative additional cost of its complete continuation. These structured credits are converted into state-local operator preferences to train a lightweight pairwise state router, which dynamically selects the next admissible operator during inference. Experiments on six reasoning benchmarks show that TreeCredit modestly improves accuracy while substantially reducing inference cost, achieving a better accuracy--cost trade-off than representative MAR methods.

Large language model (LLM) agents are increasingly capable of planning, using tools, and interacting with external environments. They are typically supported by harnesses, which manage state and coordinate multi-step execution. Graph analysis provides a promising setting for evaluating their agentic capabilities, because it requires agents to access data and execute operations in a graph environment. However, existing graph benchmarks for LLMs provide limited coverage of graph tasks and graph types, making it difficult to comprehensively evaluate LLM agents. Moreover, they typically formulate graph analysis as text-based question answering, where graph information is directly provided in the prompt, limiting the evaluation of end-to-end agentic capabilities. To address these limitations, we introduce GABench, a comprehensive benchmark for agentic graph analysis. GABench spans three graph types and covers four graph analysis task categories: graph retrieval, graph theory, graph machine learning, and graph open-ended question answering. GABench also provides 84 executable tools for accessing graph data and performing diverse graph operations. Building on these tools, we develop an agentic graph analysis task generation pipeline and construct 10,400 tasks with verifiable ground truth.Using GABench, we evaluate a range of frontier LLMs and agent harnesses. Our experiments reveal three key findings: (1) Existing LLM agents still struggle with complex graph analysis tasks. (2) Harness choice significantly affects performance, yet existing harnesses remain limited on complex graph tasks. (3) Graph analysis depends more on tool-call quality than quantity. Our findings provide practical insights into the development and evaluation of LLM agents for graph analysis.

To operate robustly in open-world environments, autonomous agents should be able to infer the behavior of unfamiliar systems through interaction alone, even in the absence of documentation. However, existing tool-use benchmarks expose semantic tool schemas in static environments, allowing agents to rely on prior knowledge rather than autonomous discovery. To address this limitation, we introduce ScrambleToolBench, an interactive terminal benchmark designed to isolate behavioral reasoning. By removing semantic cues and enforcing a continuous task curriculum, the benchmark requires agents to uncover hidden tool behaviors entirely through trial-and-error interaction. The benchmark further introduces dynamic challenges, including mapping drift, stochastic action failures, and temporal execution windows, to evaluate whether agents can revise and adapt their hypotheses as the environment changes. Our evaluation of state-of-the-art language models reveals that successful initial discovery does not translate into robust adaptation. When faced with structural changes such as mapping drift, agents fail to use deductive strategies such as cycle tracing, and instead exhibit belief inertia or fall back to exhaustive search. Increasing test-time reasoning only amplifies this expensive brute-force search rather than enabling deductive recovery. While equipping agents with persistent memory reduces compounding errors, they remain unable to efficiently infer structural changes, highlighting a gap in current agent reasoning.

Multi-agent large language model (LLM) systems can expose protected state through internal messages, tool arguments, logs, and persistent memory even when their public outputs appear innocuous. Existing privacy prompts, redaction methods, and source-level access controls restrict surface content or data access, but do not specify what a legitimately informed agent should disclose or how that disclosure may be reused downstream. We introduce Minimum-Necessary Communication (MNC), a typed semantic-declassification protocol that selects a task-sufficient disclosure from an application-authored candidate family and binds it to explicit recipient, purpose, forwarding, lifetime, logging, and memory scopes. A reference monitor enforces these scopes across subsequent operations, while a history-aware extension accounts for inference risk accumulated over repeated disclosures. Controlled semantic-join, memory, probing, and longitudinal experiments show that conventional defenses can preserve protocol-level utility while exposing substantial additional inference signal. Under identical receipt text, MNC preserves authorized delivery while blocking unauthorized forwarding, logging, durable storage, and retrieval after expiration that a text-only semantic declassifier permits. Two-backbone MAGPIE executions further show that mediated disclosures propagate through subsequent planning, tool use, coordination, and memory retrieval. These results support scope-bound semantic declassification as a practical communication boundary for private LLM-agent systems.

Model Heterogeneous Federated Learning (MHFL) addresses client-level resource heterogeneity by allowing each participant to train a personalized model architecture under a shared training objective. A prevalent paradigm, Partial Training (PT), achieves this by allowing each client to train a subnetwork of the global model. However, existing PT methods typically rely on predefined architectural templates or over-parameterized supernets, limiting fine-grained personalization and imposing substantial computational and memory overhead. We propose FedJigsaw, a novel framework that reshapes model personalization as a dynamic and decentralized model assembly problem. Instead of selecting subnetworks from a predefined supernetwork, each client constructs its model by assembling reusable modules learned from neighboring clients. At the client level, we introduce AttenAssemble to enable each participant to adaptively construct a tailored model based on local observations. To support efficient knowledge sharing under communication and privacy constraints, we design SymbioArchitect, a mechanism that allows clients to exchange granular model modules with their topological neighbors. To mitigate training instability introduced by decentralized module exchange, we design CoRe-Tune, an attention-enhanced centralized training with a decentralized execution strategy, which guides local policies to foster implicit collaboration and stabilize training dynamics, without compromising data privacy. Extensive evaluations demonstrate that FedJigsaw outperforms state-of-the-art MHFL baselines by up to 13.8% in relative accuracy while significantly shrinking cross-client performance variance, but also slashes decision-making latency and peak memory footprint compared to existing policy-driven methods.

Multimodal Large Reasoning Models (MLRMs) have achieved strong performance on tasks requiring visual understanding and multi-step inference. However, as reasoning trajectories grow, models may become less effective at using information established earlier in the context, increasing the risk of reasoning errors. Existing approaches primarily address this problem by sustaining visual grounding throughout reasoning. However, reasoning also transforms visual observations into task-specific relations, constraints, and intermediate conclusions whose influence may weaken over long trajectories. Our attribution analysis suggests that correctness is not consistently separated by image attribution alone, but is more closely associated with whether trajectories retain and integrate such reasoning-derived information across stages. Motivated by this, we introduce TRAM (TRajectory-derived Auxiliary Memory), a training-free method that augments standard decoding with an auxiliary memory pathway derived from the model's own reasoning trajectory. TRAM consolidates completed reasoning into a compact latent memory, updates it online through fast and slow recurrent streams, and feeds it back into selected decoder layers through a lightweight residual pathway. Experiments across four MLRM variants on eight benchmarks show that TRAM improves performance over vanilla decoding on mathematical, scientific, and general visual reasoning tasks without additional training.