Long-Horizon Tasks

Personal health management unfolds over repeated encounters, yet most health AI systems treat each request in isolation. We developed HealthClaw, an open-source agent architecture that updates support as a person's routines, preferences, measurements and risks change. It separates shared safety rules and medical knowledge from private longitudinal memory containing profile facts, reusable procedures and episodic traces. After each episode, induction determines what should update the profile, revise a procedure, remain episodic or be excluded. We evaluated HealthClaw with a synthetic year-long benchmark and nine 200-case biomedical tasks. Across 900 longitudinal support probes, answer accuracy increased from 0.2% with current-query prompting to 45.7% with HealthClaw, while prompt-side context exposure was 71.7% lower than with full-history prompting. In 100 privacy probes, HealthClaw produced higher privacy-aware answer quality and fewer unsafe disclosures than both baselines. Across the biomedical tasks, the mean absolute gain in the task-specific primary metric was 27.0 percentage points, and seven gains remained significant after false-discovery-rate correction. These offline benchmarks support governed, self-evolving memory for longitudinal personal health agents, although clinical effectiveness requires prospective evaluation. HealthClaw is publicly available at https://github.com/HC-Guo/HealthClaw.

One high-level instruction in long-horizon manipulation can cover several action stages. We use segmented action annotations as an intermediate representation between the full-task instruction and VLA action chunks. A progress module tracks the active stage, while the action policy receives stage information either as current-stage text or as a normalized ordinal stage index in robot state. We compare these interfaces with GR00T N1.6 on LIBERO-10 under direct fine-tuning and continuation fine-tuning from a full-task instruction baseline. Under direct fine-tuning, full-task instruction, current-stage text, and Ordinal Stage-State achieve mean success rates of 57.45%, 50.24%, and 54.36%, respectively, showing that explicit stage information does not automatically improve the policy. Under continuation, the corresponding means are 49.07%, 50.00%, and 53.75%, with Ordinal Stage-State exceeding both alternatives in all three paired runs. The observed benefit differs across interface representations and training arrangements.

Music visualization offers a powerful way to enhance listeners' understanding and experience of music by translating auditory signals into visual forms. However, most existing approaches either rely heavily on lyrics or generate flat, non-immersive videos similar to conventional music videos, which limits their ability to convey the emotional dynamics of music and provide an immersive listening experience. We propose Bring Music The Horizon, an emotion-aware pipeline for music-driven 360$^\circ$ video generation. Given an input song, our work first estimates its emotional trajectory by predicting valence-arousal values at the level of every four bars. These values are then converted into emotion-aware visual guidance using EmotiCrafter, and these guidance vectors can be manipulated by the SEGA framework, which provides fine-grained semantic control for keyframe generation. Finally, image-to-video models are applied to the generated keyframes to synthesize temporally continuous 360$^\circ$ videos for immersive music visualization. Our pipeline generates 360$^\circ$ music visualization videos that reflect the emotional progression and temporal structure of the input song. We demonstrate its capability using songs from different genres and provide qualitative comparisons with From-Sound-To-Sight, a representative audio-to-visual generation baseline, on our project page at https://etoile-et-toi-mp3.github.io/BMTH_Project_Page/.

While 3D Scene Graphs (3DSGs) provide crucial structured representations for embodied agents, conventional Ahead-of-Time, build-everything-then-filter pipelines conflict with the real-time, low-latency demands of edge platforms, inducing a perceptual saturation effect via severe observation redundancy. To resolve this, we present JITOMA (Just-In-Time On-demand Memory Activation), a closed-loop framework that unifies task reasoning, perception, and memory into a just-in-time growth process. Instead of exhaustively mapping the entire environment, JITOMA leverages a top-down task heatmap at the frontend to filter continuous observations, routing minimal streams to maintain a global foundation of low-cost, dormant anchors. Upon a cognitive query, the backend Large Language Model (LLM) parses the robotic intent to dynamically awaken task-relevant anchors, triggering resource-intensive operations -- such as dense node captioning and functional inference -- exclusively within the activated local subgraph. To evaluate these dynamic capabilities and study perceptual saturation trade-offs, we introduce JITOMA-Bench, a comprehensive suite for long-horizon multi-tasking and complex multi-step reasoning. Extensive experiments demonstrate that JITOMA substantially reduces active graph size and captioning latency, while maintaining stable processing time under long-horizon task switching.

Agent memory is a systems problem for long-horizon agents. Practical deployments require retention of task state across extended conversations, recovery of user-specific facts and preferences across sessions, and accumulation of procedural knowledge from prior outcomes. These requirements extend beyond document retrieval: a memory layer must determine which interactions become durable state, how that state is scoped, how it is retrieved under latency constraints, and how it is revised or removed over time. This report studies Oracle Agent Memory as a database-native memory substrate built on Oracle Database. Three themes organize the discussion: memory as a lifecycle spanning ingestion, extraction, consolidation, retrieval, summarization, and revision or removal; a layered architecture that separates an active memory core from a passive memory-store interface with explicit scope control across users, agents, and threads; and evaluation methodology in which downstream task accuracy is complemented by memory-centric measures such as evidence retrieval, recall, latency, and estimated token use. The report summarizes LongMemEval results, reaching 93.8% accuracy, compares Oracle Agent Memory against flat-history baselines, using about 10.7x fewer tokens, and published or reported external baselines where available, and closes with implementation-oriented appendix material covering setup, thread lifecycle, and search semantics.

Long-term memory has become a foundational capability for LLM-based agents that accompany users across extended, multi-session interactions. Existing benchmarks, however, evaluate such memory almost exclusively through downstream question answering, scoring only the correctness of a final answer. This black-box formulation conflates the heterogeneous causes of memory failure, such as missing the introduction of a relevant fact, binding an operation to the wrong target, or relying on stale values after a correction. As a result, it can credit correct answers despite their reliance on inconsistent or unsafe memory states. In this paper, we argue that, in dynamic long-horizon interactions, memory is not a static collection of facts but a lifecycle of explicit operations, including remembering, forgetting, updating, reflecting, and their compositions. We introduce MemOps, a benchmark that reformulates conversational memory as a sequence of lifecycle operations and represents each memory event with a structured trace specifying its trigger, target, scope, state transition, and supporting evidence. A controllable generation pipeline embeds these operations into long, task-oriented conversations and produces gold operation traces together with six categories of operation-level probes, evaluated under both adjacent-evidence and long-context settings. Across long-context, retrieval-based, parametric and managed-memory systems, MemOps disentangles failure modes that final-answer accuracy alone conceals, revealing that current systems remain far from uniformly reliable. For instance, session-level retrieval outperforms turn-level retrieval, and long-context models remain notably weak at reconstructing ordered memory-state trajectories. These results move long-term memory evaluation from final-answer scoring toward interpretable, operation-level diagnosis.

OpenClaw has emerged as a leading agent framework for complex task automation, yet it faces insufficient cross-platform GUI interaction support and a well-built self-evolution mechanism. These flaws limit its adaptation to diverse device ecosystems and prevent performance improvements through continuous learning from execution experience. To resolve these issues, we propose the Know Deeply, Act Perfectly paradigm for personal assistants, which holds that accumulated user interaction and task-running experience directly improve execution accuracy and efficiency, unifying cognitive comprehension and operational execution. Based on this paradigm, we introduce KnowAct-GUIClaw, a novel Know-Route-Act-Reflect framework designed to address OpenClaw's GUI manipulation deficits and break through its cross-platform and recursive self-improvement constraints. First, the host agent leverages accumulated interaction experience and task-relevant knowledge for long-horizon task decomposition and allocation (Know). Second, a pluggable GUI subagent with an experience-attributable memory system (Know) and self-evolving skill library (Act), enabling seamless cross-platform migration and fast-path integration. Especially, this framework continuously stores user profiles and feedback to improve the accuracy of task decomposition and tool calls. Extensive experiments across Android, iOS, HarmonyOS and Windows show that KnowAct-GUIClaw achieves superior efficiency, accuracy and cross-platform adaptability. Especially, the GUIClaw with open-source Kimi-2.6 models achieves the best performance (64.1%) on the long-horizon MobileWorld benchmark, beating all agentical frameworks and closed-source agentical models, e.g., Seed-2.0-Pro and GPT-5.5. Additionally, the knowledgeable memory and execution skills supported by our framework are transferable across diverse base models, improving by 8.5% with Kimi-2.6.

We investigate whether temporal hierarchy can improve LeWorldModel on long-horizon goal-conditioned control. We introduce Hi-LeWM, an extension that freezes the pretrained low-level LeWM and adds high-level planning over latent subgoals. We evaluate Hi-LeWM on PushT and Cube across increasing goal offsets. Hierarchy does not automatically improve performance: at short horizons, the best configuration uses a one-step high-level horizon, while longer horizons reveal a mismatch between the learned high-level action space and the inference-time search distribution. Experiments with true future latent subgoals show that the frozen low-level controller can execute well-aligned intermediate targets, indicating that high-level subgoal generation is the main bottleneck. Unconstrained search can select latent macro-actions that appear favorable under the learned model but produce poor control targets. Constraining search around macro-actions encoded from training trajectories, with appropriate subgoal execution timing, recovers useful hierarchical regimes, improving over flat LeWM by +11.3 percentage points at medium-range horizons and +14.7 percentage points at the longest PushT horizon. Overall, temporal abstraction can benefit compact frozen LeWM, but only when high-level search remains compatible with the low-level controller

Mirror Theory proposes that an intelligent system should be studied not only by what it represents, but by what coherent continuations it can sustain under repeated reflection. We make this claim operational through \emph{viable path entropy} (VPE), a finite-budget measure of verified continuation capacity. Given a mirror state, a rollout protocol, a verifier, and a mode map, VPE decomposes bounded capability into two parts: the probability of reaching a viable continuation and the diversity of verified continuation modes reached among successful rollouts. This paper restores the full theoretical scaffold behind the measure: intuition as local underdetermining constraint, taste as invariant-selecting pressure, reflection as taste-guided resolution of underdetermination, and geometry as the learned structure that makes future reflection stable. We then instantiate the theory in language-model reasoning experiments on GSM8K. Across Qwen2.5-Instruct models, 32 sampled rollouts per problem, and two reflection horizons, increasing the token budget from 96 to 160 substantially expands verified reachability, reduces zero-reachability, increases verified-mode entropy, and improves smoothed VPE. At 160 tokens, Qwen2.5-1.5B realizes the strongest mirror horizon among the tested models, even though Qwen2.5-3B has more parameters. This shows that mirror horizon is not parameter count, but accessible verified continuation capacity under a bounded reflection protocol. The result supports Mirror Theory as a measure-level account: capability is the structure of viable continuations made reachable, not merely one-shot accuracy or pass@k.

In this work, we introduce LongMedBench, a real-world EHR-based benchmark for long-horizon clinical decision-making. Prior evaluations of LLM-based medical agents have largely emphasized short-context knowledge QA and tool use. However, real-world medical care is inherently longitudinal, and clinicians must aggregate evidence across repeated visits, tests, and evolving treatments. Therefore, long-horizon interaction is essential for realistic assessment. LongMedBench is constructed via a reproducible pipeline that integrates MIMIC-IV admission records and clinical notes into time-series event streams and long-context memory datasets, enabling long-horizon, multi-session interactions between agents and a clinical environment. It comprises 335 patients, with 19.72 inpatient visits per patient on average and 44.91 medical events per visit. Guided by the long-horizon decision process, we propose an evaluation taxonomy with three suites: fact-based QA, temporal reasoning, and long-horizon decision-making. This taxonomy measures how agents understand and leverage historical patient information over extended horizons. Our experiments show that while recent LLMs can make good use of explicit timestamps, they have challenges in implicit time inference; The RAG and agent memory system can improve the performance of information retrieval tasks, but the performance of decision-making tasks is highly dependent on the model's immediate context.

Deployed LLM agents rely on agentic context, the model-external textual control content assembled by an operational harness. In this work, the mutable component of that context is a persistent system-level instruction that is updated from operational experience while the model, tools, and harness remain fixed. Over long evolution horizons, flat-text maintenance makes verification increasingly difficult as accumulated instructions grow and interact. We propose Graph-Regularized Agentic Context Evolution (GRACE), which maintains the persistent instruction component as a typed semantic graph and validates proposed updates within the local typed neighborhoods of modified nodes. Accepted graph updates are reconstructed as incremental edits to the textual instruction checkpoint used at deployment. We evaluate GRACE within a fixed telecom agent harness derived from $τ^2$-bench under a controlled distribution-shift protocol. Across five independent replications, GRACE improves strict reliability, measured by pass^3, from the Gemini 2.5 Flash zero-shot value of 0.091 to 0.673$\pm$0.136 at the final checkpoint. This exceeds a Gemini 3.1 Pro zero-shot reference of 0.242 on the same held-out set, while the flat-text HCE baseline finishes at 0.191$\pm$0.051. These results identify two requirements for reliable long-horizon context evolution, a structural substrate that makes verification local and a consolidation mechanism that keeps accumulated instruction content usable.

AI agents have become capable of autonomously completing short, well-specified tasks. However, existing terminal benchmarks largely focus on simple problems that finish within minutes and are evaluated only by their final outcome. This setup overlooks intermediate progress and partial solutions, yielding sparse reward signals and an incomplete picture of agent capability. We introduce Long-Horizon-Terminal-Bench, a terminal benchmark of 46 long-horizon tasks spanning nine categories, including experiment reproduction, software engineering, multimodal analysis, interactive games, and scientific computing. Each task follows a Terminal-Bench-style setup with a reference solution or simulation engine, but is further decomposed into fine-grained graded subtasks. This design enables dense intermediate rewards and partial credit, allowing evaluation to capture not only whether an agent reaches the final goal, but also how far it progresses on open-ended workflows. Tasks in Long-Horizon-Terminal-Bench typically require hundreds of episodes and minutes to hours of execution, stressing long-horizon planning, long-context management, and iterative debugging rather than one-shot problem solving. We evaluate 15 frontier models and find that agents consume on average 9.9M tokens per task, with roughly 231 episodes and 85.3 minutes of execution time per run, making Long-Horizon-Terminal-Bench more demanding than prior terminal-based benchmarks. Even the strongest tested model achieves 15.2% pass@1 at a partial-reward threshold of 0.95 and 10.9% at a perfect-reward threshold of 1.0, while the mean pass rate across models is 4.3% and 1.7% under the two thresholds, respectively. These results reveal headroom for improvement. We further analyze failure modes and error patterns, and release Long-Horizon-Terminal-Bench to support future progress on long-horizon terminal agents.

Recovering high-quality video from sparse event streams is a challenging task. Regression methods often blur textures, while existing generative models struggle with long-term stability. We propose LongE2V, a novel approach that leverages pre-trained video diffusion priors to jointly handle event-based video reconstruction, prediction, and frame interpolation. By fine-tuning a foundational video model, our approach achieves high data efficiency and superior perceptual quality. We introduce Autoregressive Unrolling and Adaptive Context Switching to mitigate temporal drift in extremely long sequences. We also propose Reencoding Alignment with Cross Residual Correction to ensure precise bidirectional consistency during frame interpolation. Furthermore, Event Voxel Density Augmentation ensures robustness across varying sensor resolutions. Extensive experiments on real-world benchmarks demonstrate that LongE2V outperforms state-of-the-art methods across all three tasks, exhibiting exceptional temporal coherence and zero-shot generalization. Project page: https://cdfan0627.github.io/LongE2V-page/

In long-horizon tasks, decision-relevant state is often scattered across an expanding trajectory, while the action agent must surface it and act. As trajectories grow, task requirements, environment facts, prior attempts, diagnoses, and open subgoals can be buried in the context window or pushed beyond it, failing to influence decisions when needed. We call this failure mode "behavioral state decay". We study memory as an active intervention mechanism rather than passive retrieval. A separate memory agent runs alongside an unmodified action agent, updating a structured memory bank from the recent trajectory and deciding whether to inject a memory-grounded reminder or remain silent. The module is plug-and-play with frontier action agents and existing agent harnesses. Across Terminal-Bench 2.0 and $τ^2$-Bench, it improves pass@1 for both weaker and stronger action agents, with gains of +8.3 pp on Terminal-Bench and +6.8 pp on $τ^2$-Bench. Ablations show that selective intervention outperforms passive bank exposure, always-on injection, advisor-only guidance, and general retrieval. As an early step toward open-weight memory policies, we train Qwen3.5-27B on SETA using SFT and GRPO, improving validation reward and achieving partial transfer to Terminal-Bench.

Long-horizon robot planning requires jointly reasoning over semantic task structure and geometric feasibility. To successfully execute a task, a robot must decompose goals, select task-relevant objects, and sequence actions, while ensuring that plans satisfy spatial constraints such as limited free space and object collisions. In this work, we propose APIVOT, a VLM-based planner that adaptively interleaves language and visual thoughts for long-horizon planning. APIVOT learns to leverage language for semantic reasoning, while using visual thoughts as imagined future states for internal verification of geometric feasibility. On long-horizon kitchen tasks, APIVOT outperforms general-purpose VLMs and prior planning frameworks, achieving the largest gains in spatially constrained settings. We find that APIVOT learns meaningful modality selection behavior, demonstrating that adaptive interleaving of vision-language thoughts improves both planning success and reasoning efficiency.

The optimization of long-horizon agents increasingly relies on reflection-based mechanisms, where a large language model (LLM) acts as an optimizer to diagnose agent failures and improve agent policies. However, real execution traces are difficult to use directly for optimization: large trace collections are often redundant and heterogeneous, making optimization inefficient and prone to overfitting to low-value failures; meanwhile, each individual trajectory also contains many irrelevant steps, while naive context reduction methods such as truncation or sliding windows can discard causally important evidence and produce misleading optimization signals. To resolve this dilemma, we introduce STRACE (Structural TRajectory Analysis and Causal Extraction), a framework that constructs high signal-noise optimization contexts for more precise and effective optimization. At the batch level, STRACE mines failure patterns to filter redundant traces and retain representative failures; within each selected trace, it performs causal localization over a textual dependency graph to remove non-causal steps and identify the true root-cause module for optimization. Empirical results demonstrate that STRACE significantly outperforms standard context-filtering baselines. Notably, on a challenging formal verification task (VeruSAGE-Bench), it successfully optimizes human-expert designed agents, delivering $1.4\times$ success-rate improvement (42.5% to 58.5%). The code is available at https://github.com/moomight/STRACE .

Large language models (LLMs) demonstrate remarkable reasoning capabilities, yet their stateless architecture fundamentally limits deployment in long-horizon research workflows requiring multi-session continuity and quantitative rigor. Here we present Ensemble QSP, a multi-agent framework featuring a three-layer hierarchical memory architecture that keeps injected context bounded and constant in project duration (mid-term project state: median 301 tokens, max 4,050, across 104 runs) by capping each state category and evicting completed work, enabling continuous autonomous operation without context degradation. The system orchestrates five specialist worker agents under domain-expert principal investigators, enforcing physical constraints through physics-based checklists and structured-domain knowledge. Comprehensive benchmarking demonstrates robust autonomous pharmacokinetic-pharmacodynamic model selection without human intervention, consistent result quality across both lower-cost and frontier LLMs, improved PK parameter recovery relative to single-agent baselines, and stable model selection across linguistically diverse prompts of the same task. Feature-level ablation across physiologically based pharmacokinetic (PBPK) models spanning a broad complexity range shows that PI-agent oversight improves debugging efficiency while preserving final accuracy across conditions. The architecture is structurally domain-agnostic, adding a new scientific domain requires only a new PI agent configuration.

Vision Language Models (VLMs) and Vision Language Action (VLA) models have shown promise in robotic control. Yet, they face significant challenges regarding explainability, generalization, and compute requirements. This paper presents a Multi-Agent System (MAS) architecture that addresses these limitations by deploying specialized agents on onboard hardware - eliminating dependence on external compute. The system controls a multi-purpose autonomous mobile manipulator in a simulated industrial warehouse, fulfilling five task categories: safety inspection, warehouse maintenance, warehouse search, package quality verification, and responding to human requests. Compact VLMs (3-20B parameters) are used throughout, with fine-tuning applied to improve package inspection accuracy. A novel "Megamind" orchestration agent mitigates context retention issues inherent to long-horizon planning with smaller models. The system was validated in a hardware-in-the-loop simulation using an AMD Ryzen(TM) AI mini PC. Results demonstrate that a fully onboard MAS architecture is a viable, cost-efficient alternative to cloud-dependent deployments, with strong potential for real-world transfer. The simulation environment has been released as open source under the Apache 2.0 licence.

Learned optimization aims to improve upon hand-designed optimizers (e.g., Adam and Muon) by meta-learning small neural network optimizers over a distribution of tasks. While recent work has greatly advanced the architectural design and inductive biases of learned optimizers (LOs), their meta-training remains biased toward short-unroll learning on particular tasks, resulting in redundant computation and leaving LOs often unable to compete with hand-designed optimizers. We introduce Efficient Long-hOrizon (ELO) learning, an efficient meta-training algorithm that (1) reallocates wasted meta-training compute to longer failure regimes, achieving efficient long-horizon learning, and (2) enforces decoupled progressive expert supervision, providing stable meta-learning signals that additionally improve the generalization of LOs. Our empirical study evaluates ELO for meta-training both element-wise and matrix-based LOs. Across downstream language modeling (GPT-2-124M/350M on FineWeb) and image classification (ViT-B/16, ResNet-50 on ImageNet-1K) tasks, ELO substantially improves the long-unroll performance and out-of-distribution generalization of the base LOs. In particular, ELO-Celo2 consistently outperforms well-tuned AdamW across all evaluated tasks, while remaining competitive with Muon on language modeling. \textit{Notably, all ELO baselines require less than 7 H100 GPU-hours for meta-training.}

Mobile manipulation requires a robot to navigate to a target object or receptacle and then perform intended manipulation. However, reaching the vicinity of the target does not guarantee a manipulation-ready base pose, a problem known as last-mile navigation. Prior methods for last-mile navigation either rely on manual pose annotation or task-specific training, limiting their scalability to open-vocabulary settings with fine-grained spatial constraints. We propose UniLM-Nav, a unified framework for zero-shot open-vocabulary last-mile navigation. UniLM-Nav decomposes last-mile navigation into view selection, task-conditioned affordance grounding, and geometry-aware base-pose reasoning, all resolved with a shared multimodal large language model (MLLM) backend. Specifically, UniLM-Nav first selects a reference view that best captures the target object or receptacle from recently collected observations. It then grounds task-relevant affordance point in the selected view and lifts the result into the robot-centric coordinate frame. Finally, conditioned on the grounded affordance, task context, and robot geometry, it infers a manipulation-ready base pose for the robot. We evaluate UniLM-Nav on the OVMM benchmark, where it outperforms the previous state-of-the-art method, MoTo, by 3.13 percentage points. Analyses show that the components of our method are crucial to final performance, and that the choice of MLLM also has a substantial effect. We further deploy UniLM-Nav on a Unitree B2 quadruped robot with a 6-DoF Unitree Z1 manipulator, validating its applicability to real-world mobile manipulation tasks.

Game worlds have traditionally been built through labor-intensive production pipelines, making them costly to develop, difficult to customization, and expensive to modify after deployment. Recent advances in video world models offer a fundamentally different paradigm. Rather than explicitly authoring every component of a virtual environment, these models autoregressively synthesize future observations conditioned on the current world state and user interactions, enabling playable worlds to be generated online. Trained on both gameplay recordings and real-world videos, they can capture diverse visual appearances and physical dynamics, opening new opportunities for interactive applications beyond gaming, including embodied intelligence. In this paper, we present \textbf{AlayaWorld}, a full-stack open-source framework for building interactive generative worlds. AlayaWorld enables open-ended real-time interaction, allowing users to freely navigate and perform diverse actions such as combat, spell casting, and monster summoning. The framework unifies the complete development-from data preparation model architecture, model training, inference acceleration, and deployment-within a modular and extensible architecture. Alongside the framework, we release reproducible pipelines, reference implementations, evaluation tools, and comprehensive documentation, establishing a practical foundation for future research and real-time applications of generative world models.

Reinforcement learning has become a promising paradigm for improving large language model (LLM) agents on long-horizon search tasks, where the agent must make a sequence of intermediate decisions before receiving a final outcome. However, existing methods still face a key limitation: the rollout budget is often allocated without explicitly assessing the utility of intermediate states. As a result, substantial computation may be spent on low-value states, even though different branches can vary drastically in their informativeness. In this paper, we propose Information Gain-based Rollout Policy Optimization (IGRPO), a policy optimization framework that treats intermediate-state informativeness as the organizing principle of rollout collection. Specifically, IGRPO performs budget-aware tree-structured rollouts by allocating expansion budget according to node-level informativeness, so that more informative branches are expanded more frequently while unpromising branches are progressively suppressed. We further demonstrate that the information gain-based rollout induces an explicit limiting teacher distribution over trajectories, which naturally yields a clear policy optimization target, thereby unifying adaptive tree-structured exploration with principled policy learning under a single framework. Experiments on seven challenging search-augmented QA benchmarks demonstrate that IGRPO consistently outperforms strong baselines under the same rollout budget constraints, validating the effectiveness of leveraging the induced teacher distribution to guide policy optimization for long-horizon search agents.

Most quadruped loco-manipulation designs trade manipulation capability against stance. A trunk-mounted arm sits high and usually carries a single arm; using the legs as manipulators lifts the manipulating leg off the ground; and even leg-mounted grippers reach two-handed tasks only by rearing onto the hind legs. This paper integrates a manipulator with a prismatic slider, two revolute joints, and a gripper into each front calf of a Unitree Go2. The two arms grasp objects at ground level and manipulate with both hands while all four feet stay planted, without rearing. With one arm carrying, the base stays free to walk. A vision-language model sequences skills from a predefined library at each skill boundary, conditioned on the head-camera image and task state, for long-horizon autonomy. In simulation, the design performs three bimanual tasks: a long-horizon cabinet task under autonomous skill selection, a cooperative two-handed lift, and an inter-arm handover.

Long-horizon failure in world models is conventionally attributed to compounding error, a generic framing that does not distinguish what kind of error compounds. We propose a kinematic-vs-dynamic reframing: world models tend to imagine kinematically rather than dynamically. We operationalize this as the imagined Kinematic-Consistency Error, a per-step diagnostic that measures how far a rollout departs from a closed-form kinematic null, paired with a perturbation protocol that tests whether iKCE responds when physical conditions cross a regime boundary. We instantiate the diagnostic on a released DreamerV3 checkpoint trained on DMC walker-walk, where imagined iKCE runs roughly two orders of magnitude above that of matched real-physics rollouts. Across a friction sweep that crosses the gait-collapse boundary, the model's iKCE stays statistically flat even as the trained policy's reward collapses through the same range, providing the kinematic-not-dynamic signature. The diagnostic distinguishes kinematic from dynamic imagination at horizons longer than the embodiment's gait period.

We present key challenges and future research directions in the security and privacy of agentic AI, based on a horizon-scanning exercise that brought together thirty leading international experts from academia, industry, and government to engage in focused discussions and collaborative exercises on the emerging risks associated with the growing agency of AI.

On-policy distillation (OPD) trains a student policy by matching a stronger teacher on the student's own trajectories, offering a promising framework for language agent training. However, its application to long-horizon agentic tasks remains insufficiently explored. We identify two key inefficiencies in vanilla agent OPD: (1) full-horizon rollouts often waste wall-clock resources on tail turns that provide weak and noisy KL supervision, and (2) trajectory-level KL objectives concentrate most of the loss on shallow tokens, leaving deeper decision turns under-trained once initial behaviors are aligned. To address these challenges, we propose TurnOPD, a turn-level budgeting strategy for efficient on-policy distillation of long-horizon agents. TurnOPD consists of two budget controllers: adaptive rollout-depth budgeting, which uses probe-based turn statistics to determine rollout length, and progressive turn-normalized loss budgeting, which gradually shifts KL weighting from token-level to turn-balanced supervision. Experiments on ALFWorld, WebShop, and Multi-Hop Search with task-specialized teacher models show that TurnOPD achieves superior validation accuracy under equal wall-clock training budgets and advances the accuracy--time frontier beyond vanilla OPD.

Long-horizon agentic LLMs are increasingly limited by finite context windows, as extended interaction trajectories can exceed the maximum context length before a task is completed. Context compaction offers a natural solution by summarizing previous interaction states and continuing the rollout under a compressed context, but incorporating compaction into reinforcement learning remains underexplored. We propose CompactionRL, a reinforcement learning strategy to train long-horizon agentic LLMs with context compaction. Our approach jointly optimizes task execution and summary generation with token-level loss normalization and cross-trajectory generalized advantage estimation. This design enables the LLM agents to learn from compacted long-horizon trajectories. We train CompactionRL on top of open models and observe consistent performance gains on agentic coding tasks. CompactionRL enables the open GLM-4.5-Air model (106B-A30B) to achieve Pass@1 scores of 66.8% on SWE-bench Verified and 24.5% on Terminal-Bench 2.0, with absolute gains of 7.0 and 3.1 points, respectively. Built upon GLM-4.7-Flash (30B-A3B), CompactionRL improves Pass@1 by 5.5 and 6.8 points, reaching 56.0% on SWE-bench Verified and 20.2% on Terminal-Bench 2.0, respectively. CompactionRL is thus deployed in the RL pipeline for training the open GLM-5.2 model (750B-A40B).

While recent Vision-Language-Action (VLA) models show promise toward generalist manipulation policies, they struggle with long-horizon tasks due to their Markovian nature-relying solely on current observations. Hierarchical dual-system methods address this but suffer from a gap between high-level planning semantics and low-level execution kinematics. We introduce Cortex, a bidirectionally aligned embodied agent framework with a customized planning interface that conveys executable and tractable subtask plans from high-level VLM to low-level VLA. Specifically, we standardize manipulation subtasks into 32 canonical skill primitives and inject tractability principles, such as representative object attributes and improved trajectory reachability, into the data generation pipeline. This enables automatic annotation of over 4k hours of open-source video data and generation of 30 hours of simulation data. We further devise an event-balanced sampling strategy to construct training data for fine-tuning the framework to better handle planning ambiguity during subtask transitions, enhanced by carefully designed harness engineering from task contexts to skill constraints during inference. Both open-loop VLM and closed-loop system evaluations demonstrate Cortex's efficacy, e.g., it outperforms monolithic baselines by 3.1% on Libero-long and 4.1% on RoboTwin. Notably, Cortex's generalist VLM enables zero-shot completion of unseen real-world long-horizon tasks, such as multi-stage chemistry experiments, by simply combining with a fine-tuned VLA-a capability infeasible through VLA fine-tuning alone.

Planning under uncertainty in continuous domains is essential for autonomous systems, yet computationally demanding. Tree-based search methods such as Monte Carlo Tree Search (MCTS) remain popular, but their branching structure can require sampling budgets that grow exponentially with lookahead depth in the worst case. From a tree perspective, continuous state or action spaces become especially challenging, since the planner must decide where to search in an infinite branching hierarchy. We propose Graph Sparse Sampling (GSS), an online planning algorithm that shares sampled futures across many candidate decisions, rather than sampling separate successors for each candidate action. This branch-free graph exposes large GPU-friendly batches, while using heuristics to focus computation. We prove finite-sample performance guarantees for GSS covering full-rank or low-rank generative simulators via smoothed backups, and discrete or sampled continuous action spaces. Under suitable overlap, regularity, and action-coverage conditions, these bounds have polynomial dependence on the planning horizon, formalizing when shared futures can avoid the exponential horizon dependence of tree-shaped sparse sampling. We demonstrate continuous-control simulations where GSS substantially outperforms tree-based planners on long horizons or achieves near-optimal performance, supporting no-branching graph planning as a complementary design principle for online control.

Long-lived AI agents require continuity across interactions, but continuity cannot be obtained by simply extending the prompt window. An agent must preserve useful prior experience, retrieve it selectively, distinguish personal context from external evidence, and revise memory when the underlying situation changes. We propose an architectural memory substrate organized along two orthogonal axes: a representational axis spanning structured records, vector representations, and graph relations; and a temporal axis spanning short-term traces, medium-term abstractions, and long-term semantic commitments. Its key design constraint is synchronized structured-vector-graph memory: structured records govern eligibility, vector representations support recall, and graph relations adjudicate support, contradiction, and supersession before gated context projection. Its central claim is that reliable personalization is a memory design problem: useful memory is structured, selectively exposed, continuously consolidated, and epistemically labeled rather than stored as undifferentiated conversation history. Beyond the framework, we instantiate MRMS as a lightweight prototype implementing structured records, vector retrieval, temporal policies, and graph-based revision. The prototype exercises the core substrate mechanisms through pre-generation memory selection, revision, boundary enforcement, and evidence attribution under controlled long-lived interaction scenarios with explicit evidence requirements.

Vision-Language-Action (VLA) models have demonstrated strong capabilities in robotic manipulation by integrating visual perception, language understanding, and robot action generation. Existing research has primarily focused on improving model architectures, training strategies, and dataset scale, while little attention has been paid to how demonstrations are collected and organized. We identify demonstration organization as a fundamental yet overlooked aspect of imitation learning, as it directly affects policy learning efficiency, training stability, and policy generalization. To address this gap, we propose a simple-to-complex structured demonstration collection strategy for VLA learning using a dual-arm robotic platform. Our approach systematically organizes data through three general principles: (i) decomposing complex manipulation tasks into progressively learnable sub-skills, (ii) standardizing the interaction environment to reduce unnecessary variability, and (iii) organizing demonstrations according to progressively increasing task complexity. This structured design enables VLA models to first acquire fundamental manipulation skills before learning increasingly complex task compositions, facilitating more effective learning of long-horizon manipulation tasks. We evaluate the proposed strategy on two representative robotic manipulation tasks: block grasping and sorting, and towel folding. Experimental results show consistent improvements in task success rate and training stability compared with the baseline method of directly collecting end-to-end complete task trajectories. These findings highlight demonstration organization as a previously underexplored but important factor in VLA learning and provide practical insights into efficient skill acquisition, scalable dataset construction, and long-horizon robotic manipulation.

Large language model agents operate in partially observable, long-horizon settings where obtaining supervision remains a major bottleneck. We address this by utilizing a source of supervision overlooked in existing post-training methods: unintended yet successful goals embedded within agent rollouts. Specifically, we introduce Hindsight Supervised Learning (HSL), where an auxiliary LLM reviews each completed trajectory and relabels it with all of the natural-language goals the agent actually achieved. HSL then pairs the trajectory with its relabeled goals and uses these pairs for additional fine-tuning. To mitigate suboptimality in the relabeled data, we propose two learning techniques for HSL, irrelevant-action masking and sample reweighting. Our experiments show that HSL is flexible and compatible with existing post-training pipelines. It improves both SFT and DPO, with larger gains on long-horizon tasks with more diverse goal spaces. Moreover, HSL is sample-efficient: on ALFWorld, it surpasses baselines trained on the full dataset while using only one quarter of the ground-truth demonstrations.

Lifelong agents need more than larger context windows and better retrieval. They need memories that can persist, evolve, and be corrected without forcing the serving stack to recompute the same history on every turn or silently reuse stale runtime state. We present PLACEMEM as a systems position on lifelong-agent memory, instantiated by an executable control-plane prototype. The central claim is that agent memory should be represented as versioned capsules that unify semantics, provenance, validity, and reusable runtime state under one correction-aware identity. In the current prototype, capsules drive prompt-level text retrieval, KV-aware routing, and cascading invalidation over live streamed backends; prospective layer-frontier replay is intentionally framed as a deeper integration agenda rather than a claimed engine feature. We describe a vLLM-first prototype with persistent capsule state, concurrency-safe invalidation, an OpenAI-compatible routing sidecar, a typed metadata contract, and a benchmark harness that measures live first-token latency, reuse, and post-correction behavior. The result is both an executable artifact that demonstrates correction-aware control-plane behavior today and a concrete roadmap for replay-aware serving integration in future lifelong-agent systems.

While current AI agents support increasingly long context windows, tool use, and skill execution for long-horizon tasks, they still require memory systems to effectively leverage historical experience. Existing memory frameworks typically rely on fixed storage, retrieval, and summarization mechanisms, which can be rigid across different tasks and often require manual tuning. To address this limitation, we propose SelfMem, a self-optimizing memory framework. Inspired by prior work on self-improving AI, we follow the principle of "teaching an agent to fish rather than giving it a fish." Instead of forcing the model to follow a predefined memory strategy or format, SelfMem provides an environment with memory tools and feedback signals that allow the agent to explore, evaluate, and refine its own memory strategy. Our results show that SelfMem consistently outperforms retrieval, compression, and agent-memory baselines on BEAM across conversation scales from 100K to 1M tokens. Compared with the strongest baseline, SelfMem improves the official score by 48.7%, 40.8%, and 41.9% at 100K, 500K, and 1M, respectively. Further question-type analysis shows broad robustness across diverse memory demands, and our optimization study shows that model-guided strategy refinement further improves performance.

Large language model (LLM) agents have shown strong decision-making capabilities in long-horizon interactive tasks, yet they still struggle to effectively leverage failed trajectories: full retries incur high interaction costs, while experience retrieval tends to dilute critical experience signals. To address this, we propose PivoARL, a self-feedback retry framework for experience exploitation in LLM agents. PivoARL identifies the pivotal erroneous turn through structured reflection and performs local retry only from the corresponding pivotal state, thereby reusing the correct prefix and reducing redundant interactions. From an information-gain perspective, we further show that pivotal retry concentrates useful experience signals near the error boundary, mitigating the signal dilution caused by state-agnostic experience utilization. Based on this insight, we design a pivotal-aware credit assignment mechanism that rewards correct prefixes while isolating erroneous suffixes, and optimize reflection quality through implicit reflection returns. We conduct a systematic evaluation on 4 agent tasks and 7 search-based QA benchmarks. Results show that PivoARL achieves significant improvements on Pass@2/3 across all tasks, with an average gain of about 11.5% over MetaRL. Moreover, benefiting from contrastive preference signals induced by pivotal turns, PivoARL also consistently improves Pass@1 on over 80% of the tasks. On Minesweeper environment, PivoARL improves over GiGPO by more than 45% and reduces interaction turns by about 42% on average compared with full-retry methods. Code is available at https://github.com/yuki-younai/PivoARL.

Current Vision-Language-Action (VLA) models excel at robotic manipulation but often struggle with non-Markovian tasks requiring long-term memory and reasoning due to their reliance on immediate observations. Existing solutions face a ''frequency-competence paradox,'' where stronger reasoning models are too slow for real-time control, while faster models lack sufficient reasoning capabilities. To resolve this architectural misalignment, we propose HiMe, a Hierarchical Embodied Memory framework that decouples embodied intelligence into a high-frequency Executor for execution, a Sentry for working memory, and a Planner for long-term strategy. We also introduce a dynamic knowledge system based on cross-modal semantic schemas and active management mechanisms, allowing robots to maintain memory plasticity through ''Add, Update, and Delete'' operations. This hierarchical design effectively balances the conflict between real-time execution and slow thinking planning, significantly improving success rates in long-horizon tasks. Experiments demonstrate that this approach not only outperforms flat memory baselines but also exhibits the novel ability to self-correct its internal knowledge based on human preferences.

Time-series forecasting supports decisions in finance, en-ergy, transportation, public health, and industrial monitoring. Recent foundation models improve transfer across forecast-ing tasks, but many depend on centralized data and Trans-former attention, which restricts their use for long, high-di-mensional, and privacy-sensitive signals. This paper presents QuantFlow, a probabilistic forecasting framework that com-bines inverted sequence embedding, bidirectional Mamba state-space decoders, quantile regression, and federated learning. Each variable is embedded over the complete ob-servation window, processed in forward and reverse direc-tions, and projected to five conditional quantiles. TSMixup expands temporal diversity through Dirichlet-weighted inter-polation while preserving sequence structure. Experiments cover cryptocurrency, traffic, electricity, Electricity Trans-former Temperature, influenza, and weather data. QuantFlow obtains mean squared errors of 0.2834 on ETTm1 and 0.2218 on Weather, and a 20-client non-IID deployment retains use-ful accuracy after three communication rounds without cen-tralizing raw records. The results indicate that selective state-space modelling is a promising basis for scalable, uncer-tainty-aware, and privacy-conscious time-series prediction, while also revealing limitations on irregular epidemiological signals and long-horizon generalization.

Long-horizon LLM agents accumulate tool results, files, plans, and user constraints that are too structured to be treated as a disposable text suffix. Current systems mostly rely on in-run heuristics such as chronological pruning and tool-output masking, or on final self-summary near a context limit. Heuristics are cheap but blind to future dependencies; summaries preserve narrative state but often hide exact evidence, locators, and editable artifacts. We present Self-GC, where GC denotes self-governing context while deliberately echoing garbage collection: the system does not merely reclaim unused tokens, but governs the lifecycle of agent context objects. Self-GC turns user turns, tool spans, and skill state into indexed objects; asks a side-channel planner to propose fold, mask, and prune actions; and lets the harness enforce recoverable sidecars, safe commit boundaries, and cache-aware commit. On a 33-session Hard Set, Self-GC prunes 43.95% of prefix tokens while leaving 84.85% of future continuations unaffected, compared with no-impact rates of 54.55% to 69.70% for heuristic baselines. On a 332-session production-derived suite, three planner backbones reach no-impact rates of 91.27% to 94.58%, while baselines remain at 77.71% to 87.46%. In production, an online account-level split reduces daytime average input tokens by 10% to 15%, with peak reductions near 20%. These results point to context management as runtime lifecycle control over indexed, recoverable objects rather than post hoc text cleanup.

Reward design remains a central bottleneck for autonomous robot policy improvement, especially in long-horizon manipulation tasks where sparse success labels provide too little signal and binary preferences collapse many competing notions of quality into one ambiguous signal. We introduce Freeform Preference Learning (FPL), a method for learning robot policies from freeform human preferences. Rather than asking annotators which of two trajectories is better overall, FPL lets them define natural-language preference axes, such as speed, safety, quality of placement, or carefulness, and provide pairwise preferences along each axis. These annotations are used to learn a language-conditioned reward model that maps a trajectory and preference label to an axis-specific reward. We use this model to train a reward-conditioned policy that optimizes across the multiple human-specified dimensions. Across four real-world and two simulated long-horizon manipulation tasks, FPL improves over sparse-reward and binary-preference methods by 38 percentage points. Beyond improved performance, FPL learns dense progress signals without explicit subtask segmentation, shows compositionality of behavior not present in the data, and allows users to steer the policy towards different behaviors at test time without retraining. Blog post with videos available at https://freeform-pl.github.io/fpl.website/

Water looks unchanged as it warms, then at a critical point it boils. We ask whether long-horizon language agents show an analogous transition in their implicit world models. In some parameter settings, changing state load by a small amount, or adding a single step of horizon, leaves behavior nearly unchanged; near a critical boundary, the same small change causes a sudden world collapse. We study this effect in a deterministic task family with exact per-step gold state. A large grid search over state cardinality, dependency density, horizon, branching, observation mode, and mutation rate reveals a phase diagram: a solved plateau, a narrow transition band, and a collapse floor. Per-step traces show the mechanism: world-state fidelity fails before action validity, so the agent is not merely choosing a bad action; it is acting from a corrupted world. Stronger models translate the critical boundary but do not remove the qualitative transition. These results make world-model collapse a measurable bottleneck for long-horizon agents.

World models offer a principled way to equip long-horizon LLM agents with foresight: predictions of action consequences before execution. However, unreliable foresight can be ignored, misused, or even degrade downstream decision-making. In this paper, we introduce WorldEvolver, a self-evolving world model framework that revises its deployment-time context while keeping the downstream agent and all model parameters frozen. WorldEvolver integrates three modules: (i) Episodic Memory, which exploits real action transitions through retrieval-based simulation; (ii) Semantic Memory, which extracts persistent heuristic rules from prediction-observation mismatches; and (iii) Selective Foresight, which filters low-confidence predictions before integrating them into agent reasoning context. We evaluate WorldEvolver on ALFWorld and ScienceWorld, measuring world model prediction accuracy on Word2World and downstream agent success rate on AgentBoard. Extensive experiments show that WorldEvolver achieves the highest prediction accuracy across three backbones and leads other world model baselines on downstream agent success rate, demonstrating that test-time memory revision enhances both predictive fidelity and planning performance.

We introduce Agents-A1, a 35B Mixture-of-Experts Agentic Model that reaches trillion-parameter-level performance by scaling the agent horizon. We investigate agent-horizon scaling from two perspectives: scaling long-horizon trajectories and scaling heterogeneous agent abilities. To support this goal, we build a long-horizon knowledge-action infrastructure that connects external knowledge, actions, observations, and verifier outcomes, producing agentic trajectories with an average length of 45K tokens. Based on this, we train Agents-A1 with a three-stage recipe. First, we perform full-domain supervised fine-tuning to align the base model with broad agentic behaviors. Second, we train domain-level teacher models to capture specialized expertise in each domain. Third, we propose a multi-teacher domain-routed on-policy distillation with salient vocabulary alignment to improve knowledge transfer efficiency across different domains, unifying six heterogeneous domains into one deployable student model. Agents-A1 achieves strong and broad performance for long-horizon agent benchmarks. Compared with 1T-parameter model such as Kimi-K2.6 and DeepSeek-V4-pro, Agents-A1 achieves leading results on SEAL-0 (56.4), IFBench (80.6), HiPhO (46.4), FrontierScience-Olympiad (79.0), and MolBench-Bind (56.8), and remains highly competitive on SciCode (44.3), HLE (47.6) and BrowseComp (75.5). We hope this work provides the community with a practical path for scaling the horizon using a 35B agent that can reach or match the performance of 1T models on long-horizon tasks.

We introduce SWE-Interact, a new testbed for evaluating coding agents on multi-turn, interactive, user-driven software engineering tasks. Existing frontier SWE benchmarks typically provide complete requirements upfront and evaluate agents on autonomous implementation. In contrast, SWE-Interact places agents in a realistic developer workflow: a carefully designed user simulator starts with vague or incomplete instructions, progressively reveals requirements, inspects the agent's workspace, and provides targeted feedback, revisions, and new constraints until the full task goal has been handed off. Grounded in large-scale studies of real coding-agent interactions, this setup tests whether agents can discover user intent, adapt to evolving requirements, and build on their own prior work. Across a suite of frontier and open-weight models, we find that strong performance on single-turn SWE tasks does not reliably transfer to multi-turn, user-driven workflows: the best-performing models solve roughly 50% of single-turn baseline tasks but only 25% of the corresponding SWE-Interact tasks. The strongest models in our evaluation, including Opus 4.8 and GPT 5.5, start strong even in the face of vague initial instructions, persevere until all the requirements are surfaced by the user, integrate them better and write clean code. However, they still suffer from over-agentic coding, forgetting requirements and technical mistakes. Weaker models start poorly under ambiguity, give up early, forget or ignore instructions and rework their code more. Overall, SWE-Interact measures an orthogonal, real-world capability axis for frontier model development: interactive goal discovery and iterative refinement with a user in the loop.

General-purpose robot policies should be modeled as dynamical systems, yet many VLA and generative imitation policies still rely on present observations or short windows. This Markovian shortcut fails in memory-dependent manipulation: identical observations can demand different actions after different histories. We present Chronos, a physics-informed full-history framework for non-Markovian long-horizon manipulation. The key idea is to elevate observation history from auxiliary context to the latent state of the policy dynamics. At each physical control step, Chronos forms one state-representative token by fusing observation and proprioception, so the token sequence is aligned one-to-one with physical time. A selective state space model propagates this causal historical state, which conditions a multimodal coarse action prior through implicit maximum likelihood estimation (IMLE). This prior is then refined by a second-order Schrodinger-inspired bridge that predicts acceleration fields, yielding smoother and more physically grounded robot motion. Across 16 simulated tasks and 4 real-world experiments, Chronos is evaluated on precision insertion, general manipulation, and memory-dependent long-horizon control. On RMBench, where success requires remembering task phase, Chronos achieves 73.6% average success, outperforming Markovian VLA baseline pi0.5 by +62.4 percentage points, a 6.6x relative gain, while using 10x fewer parameters. It also surpasses the memory VLA Mem-0 by 22.8 points while using over 30x fewer parameters. In real-world dual-arm experiments using a single RGB camera, Chronos achieves 78% average success over four tasks, including 72% on the three memory-dependent tasks, whereas pi0.5 achieves 7% overall and 0% on the memory-dependent subset. These results suggest that history should not be treated as auxiliary context, but as the latent state of the manipulation policy.

In a deep forecasting pipeline for fat-tailed financial returns at short horizons, which matters more - the backbone architecture or the output head? We compare four modern backbones (TimesNet, DLinear, N-BEATS, iTransformer) under three output heads: a point head, a single-Gaussian density head, and a Gaussian mixture density head with K=4 components. On S and P 500 monthly log-returns (1871-2023) under anchored walk-forward validation, the three heads form a strict gradient: switching from point to Gaussian improves CRPS by about 1.3 percent; switching from Gaussian to mixture adds a further about 2.4 percent. Switching between backbones, in contrast, changes CRPS by less than 1.5 percent on the point-head row and on the backbone-mean axis; density-head backbone spread is larger (up to 5.1 percent on the h=1 Gaussian row, driven by N-BEATS) but the head gradient (3.7 percentage points) still dominates. The Model Confidence Set on squared errors does not exclude any of the 12 variants at the 5 percent level: the head separates them only on distributional metrics (CRPS, pinball, coverage), not on squared error. The mixture head incremental value over a single Gaussian is largest in the highest-volatility regimes (13.9 percent in 1970s stagflation at h=12), confirming the mixture captures tail risk beyond what a unimodal Gaussian can express. The picture is horizon-dependent: the head dominates at short horizons, but at long horizons (h >= 6) the backbone re-takes the lead - an h-split we document against classical baselines (section 5.1). We conclude that on fat-tailed returns at short horizons, the head dominates the backbone, and the mixture distribution adds genuine value over a single Gaussian during crisis periods when risk-management decisions actually matter.

Long-horizon strategic planning in complex strategy games requires coordinating tightly coupled decision domains, including technology, economy, diplomacy, and military, across hundreds of turns under imperfect information. Existing LLM-based agents face three challenges in this setting. They recover little relational structure, such as distance or threat, from raw coordinate observations. They serialize the entire growing game state into every prompt and plan all domains in a single output, so the context eventually overflows and urgent domains crowd out long-term ones. The only reward is a lagged final score, which provides no progress signal within or across games.<br/><br/>We present SAGA, an LLM multi-agent framework that addresses these challenges with three mechanisms: a Map-Semantic Scene Graph that renders typed entity relations as concise per-entity text, a Tool-Augmented Planner that retrieves state on demand and routes disjoint per-domain directives to specialist controllers, and a Dual-Horizon Feedback Loop that sets intermediate goals within a game and distills causal lessons across games. On CivRealm, a benchmark built on the strategy game FreeCiv, SAGA attains the highest mean final score among six methods, the statistically strongest gains on infrastructure, and a 27% reduction in output tokens. With cross-game evolution enabled, it reaches the highest score over five successive games.

The database community has repeatedly advanced the state of the art by recognizing that new workloads demand new system architectures. We argue that long-horizon agentic tasks -- code generation, scientific discovery, hardware design -- are such a workload. These agents explore: they generate artifacts, execute tools, observe failures, branch, and repair over hundreds of steps. This search produces a structured object we call an experience graph: executable artifacts, tool outputs, rewards, sibling comparisons, and causal lineage. Yet existing agent frameworks treat this experience as disposable state -- JSON checkpoints and session logs that cannot be recovered after a crash, queried across users, or materialized into training data. We propose Trellis: a data foundation that treats the experience graph as first-class, governed, queryable database state. The core insight is that search over experience graphs is a database access pattern. Frontier selection is a query, cross-session reuse is vector-seeded graph retrieval, training-data extraction is a materialized view, and reconstructing what an agent knew at any past step is a time-travel query. When the database owns the experience graph, agents become stateless compute, and crash recovery, horizontal scaling, and a closed-loop training flywheel emerge as architectural byproducts. We ground the design in KernelEvolve, a production accelerator-kernel optimizer at Meta, where cross-session reuse reaches a target speedup roughly 10x faster at 52% lower token cost. More broadly, Trellis turns inference-time search from disposable computation into a durable institutional asset: logs made databases reliable; experience graphs may make agents cumulative.

Long-horizon embodied manipulation requires agents to remember persistent objects, track changing scene states, and reuse prior interaction knowledge. However, existing agent memories are often stored as unstructured histories or embedding-based records, making it difficult to retrieve manipulation-relevant object parts, physical states, action effects, and executable skills. We propose an analytic concept-centric memory framework for agentic embodied manipulation. Our memory organizes experience around structured analytic concepts, where objects are represented by semantic parts, parametric templates, grounded poses, affordances, and manipulation states. It further connects object and scene memories with transition memory for action-induced state changes and skill memory for template-grounded and policy-grounded execution. At runtime, the agent performs structured coarse-to-fine retrieval to identify relevant objects, states, transitions, and skills, supporting state-consistent reasoning and skill reuse. Experiments on memory-dependent manipulation, articulated-object generalization, real-world memory evaluation, and ablations show that our approach improves task completion, retrieval accuracy, object re-identification, and cross-object skill generalization over unstructured and embedding-based memory baselines.

Extensive context has become the norm as Large Language Models (LLMs) are increasingly deployed in long-horizon tasks. The concern that increasing context length degrades model capabilities, known as context rot, has become a central issue for these applications. In this paper, we focus on deep search scenarios, aiming to investigate the rot phenomenon and its mitigation strategies. By evaluating four flagship open-source models across three benchmarks, we reveal a prevalent but unnoticed rot phenomenon: extensive context causes models to directly give up or prematurely provide uncertain answers, and this issue is exacerbated as the context grows. Through pruning experiments, we demonstrate the relationship between the accumulated context and the rot phenomenon. Furthermore, we investigate mitigating this issue through context management and post-hoc rejection sampling. For context management, we systematically evaluate seven different methods across three categories, based on performance, cost, and impact on context rot, providing clear guidance for strategy selection and usage. For rejection sampling, we develop a rot-aware filtering strategy and demonstrate its effectiveness across three aggregation methods. Finally, we show that these two approaches can be combined for further performance improvements.

Existing computer-use benchmarks fail to capture the realism, complexity, and long-horizon demands of real-world computer use, limiting their ability to reveal the limitations of frontier agents. We introduce OSWorld 2.0, a benchmark of 108 long-horizon computer-use workflows across everyday and professional tasks, designed to capture complex and challenging real-world phenomena. Each task represents a realistic end-to-end workflow that takes human users a median of about 1.6 hours to complete and requires an average of 318 tool calls with Claude Opus 4.7 using maximum thinking, compared with about 30 in OSWorld 1.0. OSWorld 2.0 targets challenge phenomena that are common in real workflows yet underrepresented in prior benchmarks, spanning interaction-design challenges such as streaming interaction and dynamic environments, as well as agent-pattern challenges such as cross-source reasoning, implicit-state inference, and visual-spatial precision. Tasks are grounded in authentic input artifacts and cross-referenced against realistic stateful user profile data, and include separate safety reports auditing safety-sensitive execution. Under our primary binary-completion metric at 500 steps, Claude Opus 4.8 with maximum thinking and batched tool calls scores best but still completes only 20.6% of tasks at a 54.8% partial score; GPT-5.5 is far more token-efficient yet plateaus near 13%. These results show that current agents are still far from professional-level computer use: rather than stumbling on basic GUI control or coding, they lose track of constraints, miss information that arrives mid-task, guess rather than ask the user, and skip verification, struggling most when a task hinges on hidden state they must recover.

Multi-robot manipulation requires jointly reasoning about contact formations, robot motions under coupled dynamics, and collision avoidance. Systematically searching over this large space is difficult and becomes increasingly intractable as the number of robots grows, the task horizon lengthens, or the scene becomes more densely cluttered. Existing approaches therefore either learn to solve the problem end-to-end via reinforcement learning or restrict planning to a simpler surrogate problem, such as planning object motions while learning short-horizon contact primitives. However, neither paradigm scales to the problem instances we target: long-horizon multi-robot manipulation in extremely dense environments. In this paper, we propose Long-horizon Adaptive Manipulation Planning (LAMP), a framework combining a generative model for manipulation with classical planning for long-horizon reasoning. We instantiate our framework with two algorithms leveraging insights from established planning techniques, A* and lazy search: LAMP-A*, which systematically searches over the coupled object-robot space, and LAMP-Lazy, a lazy planner that enables real-time replanning through deferred evaluation. Experiments in challenging simulated environments demonstrate that our approach solves complex long-horizon tasks in highly cluttered environments that prior methods cannot handle.

Large language models (LLMs) are increasingly used to take actions in the real world and support human decision-making, yet most agents rely on parametric knowledge, fixed post-training data, retrieval, or search. This paradigm breaks down in novel domains and for sophisticated queries that cannot be answered from prior knowledge alone. Knowing the laws of physics, for instance, does not by itself enable LLMs to answer queries or complete long-horizon tasks in a complex physical system. To address this, we introduce Hierarchical Experimentalist Agents (HExA), an in-context self-improvement framework to learn from active experimentation. HExA iteratively designs and refines query-relevant experiments, learns a reusable library of composable skills from experience, and integrates experimental evidence to answer queries or take actions. HExA is training-free, compatible with any black-box model, and does not require external supervision, oracles, or offline data. To evaluate active experimentation, we introduce Interphyre, a tool-calling benchmark built on the PHYRE 2D procedural physics environment, where agents propose interventions and test hypotheses through simulation APIs. Experiments show that current LLM agents struggle in these settings, especially on the hardest levels of Interphyre. Claude Sonnet 4.6 achieves only 2% success, while HExA improves the same model to up to 77% success. HExA also improves open-weight models and outperforms agentic baselines such as ReAct and Reflexion. Moreover, using only skills learned from easier levels and transferred without active experimentation, HExA achieves 44% success, demonstrating the reusability and generalization of its learned skills. Overall, HExA shows that learning through active experimentation can help agents discover useful knowledge, acquire reusable skills, and make efficient progress on novel long-horizon tasks.

We introduce the Complexity Ceiling Benchmark (CCB), a controlled evaluation of how language-model reasoning decays as the number of required sequential steps grows. CCB fixes the semantic content of a task and varies only its depth N in {5,...,50} across three structurally distinct regimes: grounded spatial state-tracking, abstract symbolic pointer manipulation, and transitive relational inference. Across 6,000 trials over five frontier and open-weight LLMs we find a consistent pattern of geometric per-step decay with widely separated domain ceilings: on the first two regimes the strongest models retain pd>0.92 across N=50; on the third every model collapses by N=5, with the best model's 50%-success horizon at H0.5~4.7 steps despite pd=0.863. A trace-level metric (TFBC) shows that 14.5% of correct answers across the benchmark are reached via incorrect intermediate reasoning. Forced verbose state-tracking does not move the ceiling (McNemar p=1.000), and the mean step at which reasoning first diverges, k*, predicts within-domain accuracy better than parameter count. CCB and the geometric decay model together reduce a model's long-horizon reasoning profile to one interpretable number per task family.

Diffusion-based video relighting enables controllable relighting from a single input video, but modern video diffusion backbones are trained on short clips and applied to long-horizon videos through chunked sliding-window inference, often causing temporal discontinuities at chunk boundaries. We address this by reframing long-horizon relighting as \emph{temporally conditioned latent domain translation}. Our framework enforces cross-chunk continuity by propagating target-domain latents across boundaries and makes this behavior learnable using \emph{masked target-domain self-conditioning}, training the model to continue from temporally masked propagated context. We further introduce \emph{warm-start prompting} with a relit prompt anchor from a controllable generative model, which establishes the initial target-domain state and creates a general interface for prompt-based relighting. Experiments on in-the-wild long-horizon videos show markedly improved temporal consistency, with chunk-boundary artifacts largely reduced and unwanted appearance changes across chunks greatly suppressed.

Vision-Language-Action (VLA) models have demonstrated strong capabilities in robotic manipulation, but their performance degrades significantly in long-horizon tasks due to cumulative error propagation. This limitation largely arises from static feature fusion mechanisms that rely on fixed weights to combine visual, language, and action representations, preventing the model from adapting to different phases of task execution. To address this limitation, we propose S$^2$-VLA, a framework that introduces a State-Space Guided Adaptive Attention (SSGAA) mechanism. SSGAA maintains a belief state that tracks task progression and generates dynamic gating weights to adaptively fuse information from three complementary sources visual features for spatial perception, task intents for high-level task planning, and temporal action sequences for execution consistency. This adaptive fusion allows the model to shift its focus throughout task execution, aligning with the evolving requirements of different task stages. Despite its compact 2B parameter size, S$^2$-VLA consistently outperforms larger 7B-scale models and achieves state-of-the-art performance on long-horizon manipulation benchmarks, including LIBERO and SimplerEnv. highlighting the importance of adaptive feature fusion for long-horizon robotic manipulation.

Large language models (LLMs) have attracted widespread attention from academia and industry, yet their deployment raises critical security concerns regarding robustness and reliability. Planning, a core component of intelligent behavior, remains challenging for LLMs, which often produce infeasible or incorrect solutions in long-horizon decision-making tasks due to inherent complexity. In this paper, we propose a symbolic feedback-driven iterative self-refinement framework to enhance the robustness and reliability of LLMs in long-horizon planning. Specifically, a natural language prompting mechanism is introduced to map logical symbols into natural language descriptions, enabling LLMs to better capture task constraints and semantics. We further design a symbolic verifier that identifies errors and converts them into corrective instructions interpretable by the LLM, thereby guiding self-refinement. In addition, we leverage a plan recognizer to infer goal reachability, facilitating more effective guidance toward desired goals. Empirical results demonstrate that the proposed framework consistently improves both feasibility and correctness in long-horizon planning tasks. This highlights its effectiveness in enhancing the reliability of LLM-based planning and potential to enable more trustworthy AI systems.

Long-horizon software engineering agents often need to manage lengthy and noisy interaction histories under limited context budgets. Existing memory management methods typically rely on static compression workflows or impose rigid constraints on compression timing and granularity. Moreover, these approaches fail to jointly optimize memory management and issue resolution capabilities to improve performance while reducing token usage. We present SWE-MeM, a training framework for proactive and on-demand memory management in software engineering agents. SWE-MeM provides a flexible memory tool that lets agents decide when, what, and how to compress based on trajectory state, task progress, and remaining context budget. We train agents with synthesized proactive memory-management trajectories and Memory-aware GRPO, which jointly optimizes memory management and issue resolution through memory-aware trajectory splitting and step-level credit assignment. On SWE-Bench Verified, SWE-MeM achieves 43.4% and 60.2% resolve rate with 4B and 30B models, respectively, outperforming existing memory management baselines in both performance and efficiency.

Time-series forecasting research has been moving steadily toward larger architectures, from specialized transformers to general-purpose foundation models, on the assumption that capacity is what unlocks accuracy. We take the opposite position: most of the gap can be closed at far lower cost by tuning preprocessing rather than scaling models. We use Ridge regression as the testbed, since it has a closed-form solution and interpretable weights, which let the optimal hyperparameters be read off the search directly. We search over context length, local normalization, regularization, and augmentation on eight standard benchmarks and find three patterns. (1) Optimal lookback is strongly series-specific and often non-monotonic in forecast horizon, with fitted power-law exponents ranging from $+0.46$ on ETTm2 to $-0.19$ on Exchange and Traffic, challenging the convention that longer horizons need longer history. (2) Normalizing over a learned trailing fraction of the context, rather than its entirety, is almost universally preferred. (3) Series within the same dataset often disagree on hyperparameters; the optimal degree of cross-series sharing varies from fully shared to fully per-series. The resulting models beat prior linear forecasters on most dataset-horizon entries and exceed Transformer, MLP, and CNN baselines on six of eight benchmarks. The optimized hyperparameters also serve as a diagnostic on the data itself, revealing structures that larger models absorb silently into their learned parameters. We provide an accompanying interactive online demonstration and the code at https://sakanaai.github.io/SearchCast/.

Large language models can serve as capable long-horizon agents, but their out-of-distribution (OOD) generalization remains weak. We identify a key source of this failure as task insensitivity: when faced with similar but distinct tasks, models might apply patterns learned during training and fail to solve the task at hand. We show that models often continue with actions aligned with the original task even when the instruction is semantically corrupted and cannot be directly answered. We further find that, when we replace the task description in a trained prompt with another similar but distinct task, the model may still output the same action. This behavior is accompanied by a consistent training-time attention drift away from task tokens and toward local observations, suggesting an optimization bias toward shortcuts. To mitigate this problem, we propose Task-Perturbed NLL Optimization, a lightweight contrastive regularizer that explicitly encourages action dependence on the task instruction. Extensive evaluations show that our intervention improves task sensitivity and OOD generalization while preserving more stable attention to task tokens.

A classical intuition holds that verifying a solution is easier than producing one. For today's coding agents, this intuition is being inverted: as foundation models develop stronger reasoning capabilities and engineering harnesses grow more sophisticated, generating complex candidate solutions is no longer difficult -- reliably verifying them has become the harder problem. Every verifier we can build is only a proxy for human intent, never the intent itself. This makes verification subject to a twofold difficulty: first, intent is underspecified by nature, making it inherently hard to faithfully check whether it has been fulfilled; second, during model training, optimization widens the gap between proxy and intent -- manifesting as reward hacking or signal saturation. To address this, we characterize the quality of verification signals along three dimensions -- scalability, faithfulness, and robustness -- and argue that achieving all three simultaneously is the central challenge. We further study four reward constructions: a test verifier for general coding tasks, a rubric verifier for frontend tasks, the user as verifier for real-world agent tasks, and an automated agent verifier for long-horizon tasks. Across different task types and policy capability levels, we conduct in-depth analysis and experiments on the core challenges of reward design and how to more effectively leverage reward signals. Experiments show that targeted verification design can effectively suppress reward hacking, improve task completion quality, and achieve significant gains across multiple internal and public benchmarks. These experiences collectively point to a core observation: no fixed reward function can remain effective as policy capability continues to grow; and verification must co-evolve with the generator.