Agent 进展

Long-running tools can dominate coding-agent latency: compilers, test suites, and repository commands take seconds to minutes while the agent idles. This observation stall presents the same tension that drove out-of-order processors — asequential interface hides work that can be predicted and started early, but a speculative result may become visible only after it and every earlier step have been validated. We present TomasuLLM, a runtime that executes agent tool calls out of trajectory order while preserving task-execution correctness. It drafts future actions, runs them in isolated copy-on-write sandboxes, traces their dependencies and effects, and commits results in trajectory order only after validation against committed state. Across three benchmarks spanning sub-second to minutes-long tool calls, TomasuLLM improves the reported benchmark means and scales with tool latency: 1.31x on 100 SWE-bench Verified tasks, 1.35x on 28 Terminal-Bench 2.0 tasks, and 1.27x matched progress on 18 SWE-Marathon sessions. Across 4,010 audited commit-validation records, it produces zero false accepts.

Large language model (LLM)-based agents use Electronic Design Automation (EDA) tools to generate and revise register-transfer-level (RTL) designs under synthesis and verification feedback. Recent methods learn from this feedback by distilling reusable skills from execution traces or by training on rewards derived from EDA-tools. Both methods are typically evaluated on the tasks that produced the experience. Repeated access to benchmark feedback on the same task can reward task-specific revision rather than creating reusable knowledge that transfers. We introduce ChipMEM, a verification-grounded memory layer for EDA agents. It combines cross-task procedural memory with within-trajectory statistical guidance. Its procedural component distills and stores a skill only after it passes synthesis, simulation, or formal checks, rather than relying on model self-assessments. A Bayesian component maintains hierarchical Beta estimates over tool-call outcomes and ranks recovery strategies that succeeded under comparable errors. A common adapter applies the same memory interface to RTL optimization and testbench-generation agents while preserving each domain's tools and acceptance criteria. We measure performance on training tasks and evaluate whether learned skills transfer to unseen tasks. On RTLRewriter-Bench, under matched model and tool settings, ChipMEM produces equivalence-passing outputs on 39/54 scored designs versus 35/54 without memory; on the 49-design short suite, mean area improvement is 8.69% versus 5.66%. On held-out CVDP tasks, ChipMEM with a frozen procedural library achieves 20/20 accepted outcomes versus 18/20 without memory in a single evaluation per setting.

Group Relative Policy Optimization (GRPO) and related policy-gradient methods for training language model agents collapse an entire multi-turn rollout into a single scalar trajectory reward before it enters the policy update. When the task composes distinct skills, especially under sparse and delayed environmental feedback, this collapsing is lossy: the optimizer must implicitly infer which competency drove the outcome and how that should change behavior. We argue the right primitive is not a better scalar but a decomposition: trajectory reward should be split along subtasks before it enters the policy update. We introduce Reinforcement Learning with Decomposed Subtasks (RLDS), whose core is Subtask-Decomposed Advantage Estimation (SDAE): a replacement for the scalar GRPO advantage that splits trajectory reward into per-subtask shares on a fixed taxonomy, computes a group-relative advantage per subtask, and distributes per-token credit by weighting each subtask's advantage by its importance, concentrating it around the step where a reflection marks that subtask's execution as consequential. We evaluate on four agentic benchmarks: FrozenLake (sparse grid navigation), HotpotQA (multi-hop QA, one retrieval tool), ScienceWorld (long-horizon embodied science), and DeepResearch (long-form research, four tools, composite rubric reward). Heterogeneity diagnostics emitted during training show where decomposition pays off - gains scale with subtask heterogeneity, largest on the high-heterogeneity tasks ScienceWorld (+11.5 points, paired-bootstrap 95% CI [+9.8, +13.3]) and FrozenLake (+9.8 points, [+7.0, +12.8]), and within noise on HotpotQA and DeepResearch, where the diagnostics predicted little to recover. ScienceWorld is also more compute-efficient under RLDS than scalar GRPO (-10.9% wall-clock per step), as long rollouts amortize the fixed reflect-and-grade overhead.

With recent advancements in large language models (LLMs) and LLM-based agents, these agents are becoming increasingly autonomous and gaining broader access to act on users' behalf on the internet. However, the vulnerability of automated agents deployed on social media platforms (e.g., for managing a user's personal account) remains underexplored. Existing studies on agent poisoning typically assume that the adversary can expose poisoned content to the agent. Although such an attack is direct and effective, it is more easily detected and mitigated. In the context of social media platforms, this leaves open whether the recommendation system itself would surface such content to the agent in a more subtle manner. Through theoretical analysis, we show that the like-score mechanism used in OASIS can be exploited, and we characterize the conditions under which a multi-stage chain of poisoned posts can steer the agent's feed. Based on these insights, we further develop an algorithm that crafts realistic poisoned posts. Experiments support our theoretical findings and demonstrate the effectiveness of the proposed algorithm. Notably, by exploiting the like-score feedback loop, the attack causes the recommendation system to select poisoned posts even when their user-post similarity falls below the retrieval threshold.

Multi-agent debate (MAD) has emerged as a promising paradigm for improving the reasoning accuracy of large language models (LLMs) through iterative peer interaction. Communication topology plays a central role in this process, motivating increasingly sophisticated mechanisms that learn, adapt, or dynamically reconfigure agent interactions to improve accuracy or reasoning reliability. Meanwhile, prior studies suggest that much simpler sparse communication can already achieve competitive performance at substantially lower cost. In this work, we take a closer look at sparse MAD and ask whether complex topology control is actually necessary to improve collective reasoning. We find that a simple random-without-replacement routing policy, which lets each agent debate with two distinct and newly sampled peers at every round, provides a surprisingly strong baseline and consistently improves the accuracy-cost trade-off of sparse MAD. Building on this observation, we further study deliberation stopping and show that lightweight stopping can substantially reduce inference cost while preserving competitive accuracy. Our results suggest that sophisticated topology control such as learned topology adaption should be evaluated against strong simple routing and stopping baselines before its additional complexity is justified.

Modern language-model agents are built around the agent loop: the LLM is placed in an environment exposing a set of tools, and the LLM has full control over the workflow by alternating between tool calls and observing their output. However, certain capabilities such as long-term memory and self-improvement currently require specialized systems beyond the agent loop itself. We built an LLM agent framework, JAZ, to explore the extent to which a minimal harness that is little more than the agent loop itself can accomplish tasks these specialized systems are built for. JAZ exposes a single LLM-based primitive `invoke` and provides a set of built-in hooks that allow the programmer to apply constraints and perform monitoring. Generalizing existing code-mode agent loops, `invoke` is the simplest loop that satisfies two defining properties: (1) the LLM can write arbitrary executable code that can include recursive `invoke`; (2) everything visible to the LLM - all inputs to `invoke` as well as its interaction history with the code environment - are variables in the code environment. We motivate our design from first principles, viewing `invoke` as a language primitive representing a function whose implementation is provided at runtime by an LLM every time it is called. To validate the design of our core `invoke` primitive, we evaluate `invoke` - with only prompting, no manually designed tools, harness, or external systems (e.g., memory or the file system) - on workflows traditionally implemented through specialized harnesses. On long-horizon workflows requiring recall far beyond the context window, JAZ `invoke` outperforms Letta (MemGPT) by 8% at half its cost on the recall-heavy portion of StuLife. On continual self-improvement, JAZ `invoke` outperforms ACE by 4% at a lower cost on AppWorld.

Reinforcement learning has become a central component of large language model (LLM) post-training, yet token-level credit lacks a generally accepted mathematical definition, leaving its relationship to commonly used training signals unclear. We formulate three regularity conditions, namely Completeness, Prefix Consistency, and Neutrality, and prove that they uniquely determine token-level credit. This characterization provides a unified basis for explaining phenomena across existing algorithms and guides the development of an improved actor-critic training procedure. Through this lens, an ideal teacher in On-Policy Distillation (OPD) acts as an implicit critic, yielding an expected policy gradient proportional to that induced by token-level credit. Response-level REINFORCE Leave-One-Out (RLOO) signals match the expected policy-gradient contribution of token-level credit despite their coarser granularity. We further establish approximate credit sparsity under bounded outcome rewards and show how intermediate critic errors in Generalized Advantage Estimation (GAE) can become comparable to the underlying credit. These motivate Policy Aligned Critic Training (PACT), which adopts an Actor-then-Critic update order to apply importance sampling correction to critic training and better align the critic with the updated policy. In agentic mathematical reasoning, PACT achieves 72.87% average accuracy across four benchmarks, outperforming GRPO and PPO by 8.80 and 13.16 percentage points, respectively. On SWE-bench Verified, PACT achieves a pass rate of 67.4%, outperforming PPO, GRPO, and SAO by 2.4, 2.0, and 3.8 percentage points, respectively.

Precision medical robotics demands adaptive decision-making under strict safety, interpretability, and execution constraints. Although recent Vision-Language-Action (VLA) models show strong multimodal reasoning ability, their continuous action generation paradigm is not well suited for precision medical tasks, where reliable closed-loop operation may also depend on non-action system function calls. To address this gap, we propose MedVLA, a hierarchical framework that couples high-level multimodal reasoning with low-level function-constrained execution. We further introduce a scalable multi-agent pipeline to generate skill-oriented chain-of-thought(CoT) data for structured training. Built on different multimodal large-model backbones, MedVLA consistently improves performance after fine-tuning, demonstrating the effectiveness of the proposed framework across model variants. Under identical initial conditions, we perform 100 closed-loop flexible electrode implantation trials. The results show that MedVLA achieves a 95.0% task success rate, substantially outperforming representative VLA baselines, including OpenVLA (8%) and $π_0$ (15%), in accuracy, stability, and safety. These results indicate that structured reasoning with constrained function-level execution is a practical route toward deployable precision medical robotics.

We introduce CoDeR, a new paradigm for world modeling. Unlike existing video world models that implicitly represent world dynamics through visual observations, our system explicitly constructs an executable world with code and employs video generation models for visual realization. Specifically, we coordinate five complementary roles to translate high-level concepts into structured world rules, executable dynamics, and perceptual observations. This design enables long-term memory, open-ended interactions, autonomous world evolution, and multi-agent scenarios, where multiple entities can act, interact, and evolve persistently beyond the current observation. Extensive experiments demonstrate that our framework substantially extends the capabilities of existing world models, enabling long-term memory, open-ended interactions, autonomous evolution, and persistent multi-agent dynamics, while achieving state-of-the-art performance across multiple evaluation settings. Code and model weights will be made publicly available. Project Page: \href{https://becauseimbatman0.github.io/CoDeR}{CoDeR}.

Collaboration topology shapes both the performance and execution cost of LLM-based multi-agent systems. Because tasks differ in complexity and required capabilities, recent approaches generate task-specific collaboration graphs that specify agent participation and information flow. However, representative topology generators use either individual agents or predefined groups throughout an organization, overlooking differing collaboration needs across subtasks. Our key insight is to select granularity locally for each functional role, combining fine-grained control with reusable collaboration patterns within one organization. Learning such organizations requires exploring a combinatorial construction space with limited intermediate feedback from final-answer rewards. Therefore, we propose MAGIC, a dense-reward reinforcement learning framework for mixed-granularity graph generation. Specifically, MAGIC constructs a mixed-granularity agent graph by sequentially selecting a functional role, instantiating it as a single agent or reusable group, and connecting it to existing units. We directly optimize the construction policy using returns from trajectories sampled under the current policy and use potential-based reward shaping to provide intermediate feedback from probe-based utility and structural signals while preserving the cumulative task reward. MAGIC outperforms state-of-the-art baselines across eight benchmarks and demonstrates strong inference efficiency in our efficiency study.

Long-horizon assigned work requires an LLM agent to track the state of a task: which steps are done, blocked, cancelled, or open to repetition. Agent systems either keep this state as text in the prompt and rely on the model to read that text, or move the state into a module that enforces it, and each system is evaluated as a whole, so no one knows how much reliability comes from the state being shown, told, or enforced. We fix the task rules, the model, and paired episodes and vary how strongly task state reaches the agent: a raw transcript, an exact checklist, per-turn directives from a state machine compiled from the brief and advanced only by execution receipts, or an enforcement gate on that machine that refuses state-violating actions; every episode is scored by exact payload matching against dynamic ground truth. Across three models, two reasoning regimes, and two domains, four findings hold without per-turn reasoning: displaying accurate state is unreliable, an unverified ledger the agent writes itself beats an accurate checklist it is shown, directives help in proportion to the model's obedience, and enforcement needs no obedience but is bounded by the correctness of its state and by the matcher that maps requests to steps; per-turn reasoning at a 235B agent compresses these separations without repairing the text rungs. The same gate, compiled from $τ^2$-bench's airline policy, raises a 235B agent's pass$^1$ from 0.39 to 0.54 and changes nothing for a 35B agent that rarely violates the policy; on PM-Bench, where acting turns on recognizing a cue rather than on state, showing the record is the best rung--matching or beating both gates and reversing the ledger-over-checklist finding--and enforcing the matcher's judgement drops a 35B agent below its raw transcript. Enforcement pays when failures are state-decidable and frequent, and hurts when the gate's judgement is wrong.

Approval mechanisms have become a primary safeguard for consequential actions in LLM-agent software. Yet the action shown for approval is often not the object ultimately consumed: workflow reload, transcript projection, argument rebinding, and durable-state lookup may reconstruct it before execution. Existing fieldflow and check-coverage analyses can establish that expected fields were inspected, but not that the inspected object version reaches the sink or that no replacement intervenes between check and use. Consequently, a locally complete repair may still leave a residual authorization bypass after reconstruction. We address this problem through three designs. (1) We formulate reconstruction-stable authorization (ReSA) over representation transitions, sink dependencies, consumed versions, and grant scope. (2) We derive obligations that judge candidate repairs and expose residual sink suffixes. (3) We implement the analysis in APAS-Finder and freeze predictions before an independent sink oracle observes execution. Predictions agree with all 28 controlled outcomes. Four matched pairs require opposite judgments despite identical field-flow and check coverage; a CodeQL composite recovers all eight when supplied object flow, dominance, interference, and the same grant contract. Two analysts agree on all four model-validation dispositions and 19/20 sink dependencies. On released consumers, five repairs prevent 60/60 tested out-of-scope effects, while three mechanism contrasts expose the predicted residual effects. Repair sufficiency therefore depends on the reconstructed action consumed, not merely on an approval record or earlier checked representation.

LLM agents often use generative models for bounded decisions, raising the question of when these decisions can be handled more efficiently without reducing task success. We study REFLEX, an agent architecture that uses Jev as a fast, typed decision layer and calls a strong LLM when confidence is low, or generation is required. On a frozen 100-task benchmark, REFLEX achieves 95% success with 72.7% fewer strong-model calls than a strong-only agent, with reductions persisting across three fallback families. Controlled interventions show that reliability depends on action-set size and near-valid alternatives near authorization boundaries. External BFCL and $τ$-style evaluations reveal limited advantages over a cheap generative cascade when ordinary routing is already highly accurate. These findings identify when selective control with Jev can reduce computation and where its benefits are limited.

Soccer tactics are interactive: an attacking action changes the opponent's defensive problem, and the observed response depends on the multi-agent match state. We introduce SambaGraph, an action--reaction spatio-temporal graph dataset and benchmark for soccer tactical response modeling. From tracking and event data for all 64 matches of the 2022 FIFA World Cup, we curate 4,070 action-centered episodes represented as temporally aligned 23-node player--ball graph sequences with attack/defense views, response labels, and 26,270 split-safe attack--defense pairs. We study three questions: whether observed responses can be classified from graph episodes, whether successful defenses can be retrieved for a query attack, and whether graph-derived summaries support grounded LLM reasoning. A compact signature MLP obtains $0.796\pm0.007$ macro-F1 for response classification, while a fused graph--signature dual encoder reaches $0.471\pm0.029$ Hit@5 and $0.655\pm0.051$ Hit@10 for full-bank defensive retrieval. Hard negatives maximize pair discrimination but not retrieval quality. Local LLMs underperform supervised encoders for direct classification and do not improve over a strong original order in eight-candidate reranking, but they provide grounded tactical rationales. These results position SambaGraph as a reproducible benchmark for graph-based soccer strategy-response research. Code and dataset are available at: https://github.com/areyesan/SambaGraph.

Ontology generation from Competency Questions (CQs) is a central yet labor-intensive phase of Ontology Engineering. While large language models (LLMs) offer promising automation capabilities, current evaluations remain fragmented. Task formulations are heterogeneous, gold standards often lack fine-grained CQ provenance, metrics conflate lexical overlap with structural and logical adequacy, and reference ontologies are not always explicitly designed around the evaluation CQs. Here, we address these limitations with CQ4OE, a benchmark for the systematic and reproducible evaluation of LLM-based ontology generation from CQs. For each ontology in the benchmark, we build a CQ-driven gold OWL ontology with explicit provenance linking each CQ to the classes, properties, and axioms required to answer it. From this resource, we define two complementary evaluation tasks. CQ2Term supports term-level evaluation of CQ-specific class and property prediction over 99 CQs, and CQ2Onto supports ontology-level evaluation over 118 CQs, including hierarchy, property modeling, and axiom-level structure. We demonstrate CQ4OE with experiments using nine LLMs under zero-shot, iterative, and multi-agent generation strategies, showing that LLMs recover explicit vocabulary terms more reliably than creating ontologies, particularly in property modeling, hierarchy construction, and axiom generation.

AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self-improvement. Its significance lies in a long-standing trend, in which increased cumulative spending on R&D yields diminishing returns. Sustained self-improvement offers a way to counter this trend. We present AIDE^2, a system that implements this loop for a frontier AI research agent. It proposes changes to its own code, benchmarks modified versions of itself on a suite of AI R&D tasks, and keeps the changes that perform best on hidden evaluations. In an autonomous 8-day run, AIDE^2 discovered seven successive improvements, ranging from a new search policy to memory mechanisms that compress and manage the agent's growing context. These gains generalize to four held-out benchmarks spanning machine learning engineering, heuristic algorithm engineering, and physics-based weather forecasting, the last of which is out of distribution from the selection tasks. On all four, the strongest discovered agent matches or exceeds a human-engineered production research agent that ranks among the strongest on FML-Bench. On a separate held-out task family, the discovered agents also exhibit reduced reward hacking, a property the loop never explicitly optimized for: the rate falls from 55% to 32% during the run, 7 percentage points below the human-engineered agent. Together, these results show that an AI research agent can improve its own research efficiency through recursive self-improvement, and that these gains transfer to tasks and domains the loop never encountered.

Predicting early outcomes based on trajectory can decrease the expenses associated with agent evaluation by terminating a run once the outcome becomes sufficiently predictable, assuming that the predictor's confidence is properly calibrated. Calibration is at risk when a predictor is applied to an agent on which it was never trained, but it is not known whether such transfer failures are broad across agent systems or concentrated in specific target agent/head combinations. Using public SWE-bench Verified trajectories and a frozen dual-head early-outcome prediction pipeline, we ran a leave-one-agent-out calibration audit, a shared-predictor leave-two-agents-out control, oracle prior correction, and a robustness battery over training cohorts, task resampling, task halves, jackknife, and thresholds. Fixed-scaffold TerminalBench analysis served as a pre-registered boundary test. Broad same-predictor pairwise heterogeneity was not supported; the median pairwise corrected-gap differences were 0.0180 (SUCCESS head, 45 pairs) and 0.0385 (FAILURE head, 35 pairs), and the pre-registered heterogeneity criterion was not met on either head. Two specific combinations, gpt-5-mini/SUCCESS and claude-opus-4.6/FAILURE, showed persistent calibration-transfer errors (median corrected gaps 0.1377 and 0.1107) without a sign reversal under any frozen control. TerminalBench did not establish cross-benchmark replication: the success target produced zero decisions (INDETERMINATE), and the failure target did not satisfy the pre-registered persistence criterion. Therefore, a strong target-specific calibration-transfer error can exist within one frozen environment, but the evidence does not establish that the error is intrinsic to the model or general across benchmarks.

Learned world models are becoming essential to general-purpose agents: by predicting action consequences, they support planning and decision-making while reducing reliance on costly trial and error. This reliance creates a fundamental ambiguity: when a world-model-guided decision fails, the trajectory alone may not reveal whether the agent's decision rule or the world model caused the loss. We formalize this failure-attribution problem as a counterfactual decomposition of return loss and prove that its components are not identifiable from passive interaction, even for finite-horizon planners. This obstruction motivates Dual-Frontier, a learning principle that admits a world-model-guided decision only when its predicted advantage exceeds a certified bound on decision-relevant world-model error; otherwise, evidence is allocated to world-model verification. Action-conditioned value bounds and a closed-loop extension guarantee non-decreasing return for admitted decisions. Calibrated gates and simultaneous confidence sequences support adaptive evidence reuse, with sufficient and necessary verification bounds. Controlled learned-model experiments validate the predicted failure modes and certification behavior, while cross-backbone tool-use benchmarks instantiate the same verify-then-promote rule in realistic agent world-model pipelines, consistently improving decision quality and reliability.

Graphical User Interface (GUI) Agents autonomously interact with software to fulfill user requests, where GUI navigation stands out as the most critical and challenging capability. Mastering this capability demands a complex synergy of step-wise decision-making, state-action alignment, and long-horizon planning. While directly mixing these corresponding navigation tasks seems intuitive to simultaneously acquire these skills, such a direct combination is severely bottlenecked by inconsistent optimization objectives and profound data heterogeneity. To overcome these barriers, we propose the MaP (stands for ``Masked Trajectory Prediction''), a unified framework that seamlessly harmonizes divergent GUI navigation tasks. By modeling multi-turn GUI interactions as a trajectory and defining training objectives through component masking and prediction, MaP shifts the optimization from task-specific marginal distributions to a consistent objective. Furthermore, to handle the data heterogeneity across multiple navigation tasks, we design a role-aware adapter learning module that dynamically routes each token to a specialized representation space. Extensive experiments on five representative GUI navigation benchmarks demonstrate that MaP effectively mitigates gradient conflicts and significantly outperforms the direct mixture training, establishing a robust paradigm for multi-task GUI navigation.

LLM agents increasingly work on long-horizon tasks, and the decisions they make along the way, such as which hypothesis to test or which implementation to build on, determine the outcome of the whole run. Making these decisions well is becoming a key capability for both engineering and research agents. We refer to the ability to make good long-horizon decisions as the taste of an agent. While existing benchmarks measure the end-to-end success of agents on long-horizon tasks, none of them measures the taste of an agent. To address this problem, we build Taste-Bench, a benchmark of taste questions constructed automatically from trajectories that agents produced in engineering and research tasks. Each question presents a decision fork, a point in a trajectory where multiple directions are available and one of them leads to a better outcome, and the evaluated model chooses among these directions without seeing what happens after the fork. We mine these forks automatically from parallel attempts at the same task and from detours inside a single trajectory, without needing human annotation. We evaluate frontier models on Taste-Bench and find that the best model answers only 59.7% of the questions correctly. We further find that forks whose deciding evidence appears later in the trajectory are much harder for every model, and that a larger reasoning budget does not improve the accuracy. Finally, we show that taste can be trained. We distill the judgment of a teacher that has seen the outcome into a student model, and the student makes better decisions on unseen tasks and improves end-to-end success on held-out SWE-bench Pro tasks.

Scientific investigations into microbial natural products (NPs) present significant challenges for novices, largely due to the complexity of microbial systems, biochemical diversity, technical skill requirements, and the demands of bioinformatics and data analysis processes. To address these issues, we introduce ChatT2, a large language model (LLM)-based agent that is specifically tailored to the unique characteristics of bacterial type II polyketides. These polyketides form a structurally distinct and therapeutically important NP family. ChatT2 was developed within an autonomous multiagent framework composed of a mentor, an executor, and an evaluator, each with defined responsibilities. The mentor acts as an intermediary between ChatT2 and the user, utilizing chain-of-thought prompting to refine the intent of the user. Under the guidance of the mentor, the executor synthesizes multimodal information via retrieval-augmented generation techniques and seamlessly integrates bioinformatics and cheminformatics tools. The evaluator ultimately assesses the output of the executor to ensure the richness and accuracy of the retrieved information. Our research highlights how ChatT2, designed with this multiagent framework, addresses the challenges faced by general LLMs in terms of understanding limited, specialized corpora and complex biological information and provides both experts and novices with a valuable tool for exploring various NPs of interest. The ChatT2 webserver can be accessed at https://chatt2.site/#/chat.

Large language model (LLM) agents often handle streams of related tasks, yet standard harnesses repeatedly ask the model to reconstruct the same control decisions inside each task's context. We study whether task feedback can instead turn recurring control into reusable executable code, while reserving LLM calls for task-specific semantic reasoning. We introduce Growing Harness, a failure-guided training paradigm that learns the agent harness itself from a strategy-free scaffold that exposes fixed model and tool interfaces but encodes no task-solving controller. Function-level execution traces localize each failure to a bounded code surface, an optimizer repairs a window of failures jointly, and a success-first held-out gate rolls back repair sequences that harm prior capability. Accepted edits accumulate in one shared harness, allowing its control structure to emerge from task feedback. Across BrowseComp-Plus and WebArena-Verified with three deployment models from 4B to 120B parameters, Growing Harness achieves the highest mean success in five of six benchmark-model settings and trails the best mean by 0.7 pp. in the sixth. Relative to a Tool-Calling agent, it reduces LLM calls by 76.0-91.8% and deployed-agent inference cost by 74.4-98.6%. On WebArena-Verified, its success remains 44.7-45.3% across model scales, whereas Tool-Calling falls to 6.7% with the 4B model. Ablations show that trace-local edits, joint repair, and gate-based rollback each improve final success. These results show that persistent program growth can move recurring control out of model context and into low-cost code, yielding reusable specialist agents that remain effective with smaller deployment models.

We introduce Qwen3.8-Omni-Flash, a natively multimodal agentic model for real-world multimodal productivity. Compared with previous omni models, which primarily emphasized perception and interaction, Qwen3.8-Omni-Flash substantially improves multimodal understanding and reasoning, as well as performance on long-horizon agentic tasks. These capabilities are supported by a native multimodal co-training strategy that preserves strong text-domain capabilities while facilitating the transfer of agentic capabilities from text to audio and video tasks. The model inherits the sparse mixture-of-experts (MoE) architecture of Qwen3.8-Next and extends the context window to one million tokens, supporting long-context multimodal reasoning and long-horizon planning. These advances enable integration into production workflows as a primary agent or a specialized sub-agent, supporting video editing, long-form audio and video translation, music-conditioned music video or movie generation, and video-based note or omni-skill creation. To address the lack of native audio and video support in existing agent harnesses, we release Qwen-MM-Plugins, a lightweight open-source plugin framework for multimodal productivity. We further frame real-time multimodal interaction as a system-level challenge requiring orchestration of context and memory management, tool use, and sub-agent delegation. Accordingly, we release Qwen-Live-Harness, an open-source framework for building responsive, real-time multimodal agents based on Qwen3.8-Omni-Flash. Extensive evaluations demonstrate that Qwen3.8-Omni-Flash achieves strong performance across multimodal understanding, reasoning, long-horizon agentic execution, and video productivity tasks. These results and the accompanying open-source tools support Qwen3.8-Omni-Flash as a practical foundation for deploying natively multimodal agents in research and production.

Persuasive video generation (PVG) is a valuable yet under-explored research topic. Despite the significant advances in multimodal content generation, AI-empowered automated creation of human-made-like videos with substantial persuasiveness remains a formidable challenge. In this paper, we propose CogenPVG, a novel Cognitive-Enhanced reflective multi-agent framework tailored for Persuasive Video Generation task. Given the topic and stance from the user, we decouple the sophisticated generation process into four sequential stages: argument reasoning, storyboard planning, asset creation, and post-editing, imitating the workflow of human video producers. To ensure high persuasiveness, each stage is equipped with a pair of generator and critic agents, following a reflective refinement scheme grounded in a solid psychological theory of persuasion, the Elaboration Likelihood Model (ELM). In the argument reasoning stage, we generate highly logical and credible reasoning thoughts under the guidance of critical thinking theory, enabling cognitive enhancement via the central route of the ELM. For the other three stages, we generate and optimize multimodal assets, assembling them into a persuasive video guided by theories of heuristics, as the peripheral route of the ELM. To the best of our knowledge, CogenPVG is the first work focused on general persuasive topics, without being confined to commercial purposes. Extensive experiments and comprehensive analysis demonstrate that our framework achieves the best persuasion performance, thereby proving the effectiveness of our proposed multi-agent framework for the PVG task.

Large language model (LLM) serving is increasingly agentic, with multi-turn sessions that idle between actions yet must retain their full context. Limited GPU memory capacity forces inactive KV states to be evicted, so resuming a session incurs either costly recomputation or slow interconnect transfers. To address this, high bandwidth flash (HBF)-an on-package 3D-NAND memory offering orders-of-magnitude greater capacity than high bandwidth memory (HBM) at comparable read bandwidth-has emerged as a strong candidate. However, its high read energy and limited write endurance make it impractical to serve all KV traffic. Fortunately, our analysis shows that agentic KV states exhibit distinct access patterns: a small hot set is read for every decoding step, while a large cold pool is read only when a paused session resumes. Exploiting this, we place the hot set in HBM and the cold pool in HBF, forming a hot-cold KV hierarchy within the GPU memory tier. On agentic workloads with Qwen3-Coder-30B-A3B, our design delivers 14 ms time-between-tokens (TBT) and adds only $\approx$0.1 ms of resume latency on top of prefill, while hosting $24\times$ more concurrent sessions per GPU. By confining HBM to the hot set, our design also cuts read power by 7.6 kW per 8-GPU node relative to serving all KV from flash-establishing HBF as a cold-tier complement to HBM rather than its replacement.

Recent advances in large language models (LLMs) are driving the emergence of multi-modal and agentic services for mobile users through cloud and edge infrastructures, where long-context workloads pose daunting challenges for inference latency. Existing disaggregated LLM serving systems largely rely on hardware profiling, configuration enumeration, or heuristic scheduling, offering limited analytical guidance for cost-efficient resource allocation. In this paper, we propose SARA, a Service level objectives (SLOs)-Aware Resource Allocation framework for disaggregated agentic LLM serving systems, which maximizes goodput under a deployment cost constraint and a series of quantile-based SLO constraints. By capitalizing on queuing theory, we first model the prefill, KV cache transfer, and decode stages as an M/G/k queue, an M/G/1 queue, and a generalized birth-death process, respectively. The analysis reveals that the prefill and decode stages are dominantly limited by computational capacity and high-bandwidth memory (HBM) resources, respectively. With these mathematical models, we further derive tractable tail behaviors of different stage-wise service level metrics for both light- and heavy-tailed workloads. These characterizations explicitly map workload, model architecture, and hardware parameters to stage-wise SLO constraints and minimum resource requirements. Finally, we develop an effective resource allocation framework to maximize system goodput under limited cost budgets. Simulation and hardware results demonstrate that the proposed framework accurately predicts the stage-wise SLO with mean errors below 5%, and improves system goodput by 26.6% on average over state-of-the-art baseline methods under the same deployment cost.

The performance of large language model (LLM) workflows often depends on selecting a small set of in-context demonstrations to guide model behavior on new tasks. Recent methods improve this process by augmenting prompts with successful reasoning paths. However, their demonstration selection relies on random sampling or metric-based rankings, overlooking the semantic structure of the task. We propose ClusterFewshot, a strategy that combines semantic structuring with utility-aware scoring to construct representative and effective few-shot demonstration sets. Evaluated within DSPy-based pipelines, ClusterFewshot substantially reduces optimization cost across multiple benchmarks, while consistently improving accuracy relative to prior bootstrap-based methods in both standalone prompt tuning and hybrid prompt-weight optimization.

Wearable health questions often combine data retrieval, longitudinal analysis, and health advice over structured records. Prompting a single large language model with a complete record and a composite query obscures whether every request is executed and which evidence supports the answer. We propose a task-oriented multi-agent framework that represents a composite query as distinct intents and typed tasks with explicit intra-intent dependencies. Specialized agents execute retrieval, analysis, and advice tasks; isolated intent states preserve request boundaries and evidence relationships before aggregation. We evaluate the framework on a synthetic dataset of $10{,}000$ virtual users with one month of longitudinal wearable records, covering structured data retrieval, multi-intent recognition, and overall response quality. Across $1{,}500$ retrieval questions, the Query Agent achieves $98.3%$ accuracy, compared with $97.9%$ for the Direct LLM baseline, while reducing average query-stage token consumption from $6{,}869$ to $3{,}136$. On $180$ multi-intent questions, the Manager Agent achieves $100.0%$ Multi-Intent Coverage and $94.4%$ Multiset Jaccard Similarity. Under the current synthetic evaluation setting, our method receives higher mean Trustworthiness and Transparency scores on both question categories, whereas Actionability does not improve consistently. These results provide preliminary evidence that explicit task organization can support task-relevant data access and data-grounded longitudinal analysis, while leaving health advice generation and validation on real wearable data as open challenges.

We present Qwen-Audio-Agent, a harness that combines full-duplex voice interaction with asynchronous task execution through a foreground-background architecture. A Frontend Agent manages dialogue and selects between direct tool use and delegation, while a Backend Agent carries out delegated tasks in a separate context. An Orchestration Runtime maintains task state, coordinates requests for user input and authorization, and schedules the return of results to the conversation. The runtime separates speech interruption from task cancellation and execution completion from result delivery, allowing conversation to continue while delegated work proceeds. Environmental events and persistent memory provide context within and across sessions. Independent adapters support integration with different frontend models, backend agents, and clients. We instantiate the architecture in desktop assistance, intelligent cockpits, and voice customer service. On an in-house cockpit benchmark of 134 cases, mixed execution achieves a task success rate of 91.04%, compared with 72.39% and 80.60% for the direct and all delegated configurations, respectively. In a separate latency evaluation on matched successful turns, mixed execution reduces mean task execution latency by 26.73% and 30.91% relative to these baselines, respectively. These results support the complementary use of direct tool calls for immediate operations and backend delegation for multi-step tasks.

Because significant action to counter global warming requires massive public support, it is important to understand the dynamics of public opinion on climate issues. Of special interest are social tipping points, as revealed by large-scale effects of small perturbations in individual behaviors. Agent-based models (ABM) are an effective computational tool for studying these matters, because they allow controlled and systematic exploration of the effects of interventions that may be infeasible in real-world social systems. Large language models (LLMs) have been used to endow model agents with the ability to communicate in natural language (rather than by exchanging predefined messages), as well as with personality (in the form of a narrative self and episodic memory). We leverage LLM-powered ABM to look for tipping points in the social dynamics of a micro-society in which some of the discussions are about climate change. Our agents' stance was defined by two variables: the strength of conviction about the urgency of climate action and the degree of trust in existing institutions. We quantified shifts in agents' "beliefs" by monitoring, across multiple rounds of conversations, (1) inter-agent distances in this two-dimensional stance space and (2) the patterns of discussion topics as modeled by Latent Dirichlet Allocation (LDA). Our findings to date suggest that significant abrupt changes in climate-change stance do occur in this simple model. We report a number of methodological lessons from this study, notably, the need to prevent LLM biases from interfering with the conversational dynamics and, more generally, to maintain agent personality and episodic memories of interactions in the face of such biases. Resolving these issues may allow for using ABM-derived insights in designing real-life interventions vis-a-vis climate change and other important societal challenges.