NeurIPS 2026 论文列表

Open agent systems (OASYS) are increasingly prevalent in real-world domains where the sets of agents and tasks change unpredictably over time. Such openness, including agent openness (AO) and task openness (TO), poses a fundamental challenge to multi-agent reinforcement learning (MARL), which typically assumes fixed state and action spaces. Existing methods address openness only partially: padding and masking approaches introduce artificial bounds, while recent graph-based or hypergraph methods handle one dimension of openness but still depend on restrictive assumptions. In this paper, we introduce Pointer Learner for Agent and Task Openness (PLATO), a pointer-network-based actor combined with a centralized graph neural network (GNN) critic, trained with multi-agent proximal policy optimization under a centralized training and decentralized execution paradigm. Our pointer-based actor outputs distributions directly over the current task set. This directly supports changing action spaces without masking or retraining. Our GNN critic encodes agent–task interactions as a graph that changes shape with task and agent composition. Together, these components consider AO and TO without the boundedness of existing approaches. We formalize PLATO in a Task-and-Agent-Open Markov Game (TaAgO-MG), extending prior task-open formulations, and prove it is well-defined over the resulting unbounded state and action spaces. We evaluate PLATO with the Methods for Open Agent Systems Evaluation Initiative (MOASEI) wildfire suppression domain, an environment designed for open multi-agent system evaluation, and we demonstrate strong performance and more consistent zero-shot generalization than state-of-the-art baselines in OASYS.

Streaming long-video understanding is a realistic setting for always-on video assistants, where visual input arrives continuously and users may ask questions at arbitrary times. We study streaming long-video question understanding (QA) under a unified temporal formulation in which multiple questions are revealed over time and may concern the past, present, or future relative to their query times. The model receives no task-type labels and must operate under strict sequential access and bounded shared memory, deciding whether each active question is already answerable or should remain open for future evidence. This differs from prior streaming settings that often focus on short streams, prefix-answerable queries, separated temporal categories, or response timing without long-horizon shared memory. To evaluate this formulation, we propose Ref2Stream (Reference-to-Stream), a general benchmark-conversion protocol that turns temporally grounded offline video QA benchmarks into streaming QA benchmarks. Instantiated on LVBench, Ref2Stream yields LVBench-Ref2Stream, which mixes past, present, and future questions under strict sequential access and evaluates both answer accuracy and answer latency. Existing streaming video methods typically improve efficiency or response timing through predefined compression, retrieval, or response policies, while offline video agents perform adaptive evidence gathering but assume access to the full video and target question. We introduce PIVOT (Planning over Incremental Video Observations for Timely Answering), a unified agentic framework that incrementally updates bounded memory from the observed stream and uses a planner-controlled action loop to retrieve relevant memory, analyze current evidence, reflect on candidate answers, and decide whether to answer or defer. Our work suggests that agentic reasoning is a promising direction for building streaming long-video systems that can adaptively gather evidence and answer only when sufficient information has been observed.

Large language model (LLM) agents increasingly operate over long and recurring external contexts, like document corpora and code repositories. Across invocations, existing approaches preserve either the agent's trajectory, passive access to raw material, or task-level strategies. None of them preserves what we argue is most needed for repeated same-context workloads: reusable orientation knowledge (e.g., what the context contains, how it is organized, and which entities, constants, and schemas have historically been useful) about the recurring context itself. We introduce **PEEK**, a system that caches and maintains this orientation knowledge as a context map: a small, constant-sized artifact in the agent's prompt that gives it a persistent peek into the external context. The map is maintained by a programmable cache policy with three modules: a Distiller that extracts transferable knowledge from inference-time signals, a Cartographer that translates it into structured edits, and a priority-based Evictor that enforces a fixed token budget. On long-context reasoning and information aggregation, PEEK improves over strong baselines by 6.3–34.0\% while using 93–145 fewer iterations and incurring 1.7–5.8$\times$ lower cost than the state-of-the-art prompt-learning framework, ACE. On context learning, PEEK improves solving rate and rubric accuracy by 6.0–14.0\% and 7.8–12.1\%, respectively, at 1.4$\times$ lower cost than ACE. These gains generalize across LMs and agent architectures, including OpenAI Codex, a production-grade coding agent. Together, these results show that a context map helps long-context LLM agents interact with recurring external contexts more accurately and efficiently.

Photonic crystal fibres (PCFs) underpin critical applications in telecommunications, sensing, and high-power laser delivery, yet their design remains a manual, expert-intensive process that couples geometric intuition with deep waveguide physics. Vision-language models (VLMs) offer a promising path toward automating this pipeline, but can they truly reason about photonic physics from microscopy images, or do they merely recognise visual patterns? We introduce PCFBench, the first comprehensive benchmark for evaluating VLMs on the complete photonic inverse-design workflow. PCFBench spans 26 tasks organised into six categories of increasing cognitive demand (Geometry Perception, Physics Understanding, Text Generation, Multi-modal Reasoning, Inverse Design, and Code Generation), built from over 220K FDTD simulation samples across 18 structurally diverse fibre families. We evaluate both individual VLMs on isolated tasks and multi-agent systems (MAS) on the full sequential pipeline from perception to design. Our results reveal a striking perception-design gap: current models can identify fibre geometry with moderate success, yet consistently fail to translate that visual understanding into physics-aware design actions. Moreover, cascading agents through the pipeline exposes significant error propagation, where upstream perception mistakes compound into downstream design failures. These findings suggest that neither model scaling nor naive agent chaining suffices; closing the loop demands explicit physical grounding. PCFBench provides the community with a rigorous, multi-dimensional testbed to measure progress toward AI-assisted photonic engineering.

LLM-based agents have made rapid progress on software engineering benchmarks, and executable evaluation is extending to chip-level hardware tasks. Board-level PCB design and upgrade has yet to receive comparable evaluation, largely because open-source PCB projects rarely provide the structured issue/PR/test histories that existing benchmarks rely on. We introduce PCBInnoBench, the first benchmark that evaluates LLM agents on executable PCB design-and-upgrade workflows. PCBInnoBench contains 372 expert-designed KiCad tasks grounded in real PCB projects and end-to-end engineering workflows. Each task pairs an agent-visible package (KiCad project and multimodal engineering references) with a hidden evaluation package (ground-truth design and expert-reviewed validators) used only for scoring. Agents must modify existing ECAD projects and pass structural, task-specific, and cross-artifact engineering-constraint checks. Each task is evaluated under two prompts of different guidance levels: one supplies an expert execution trajectory of ECAD edits together with the consulted references, the other states only the task objective, requiring agents to navigate the full workflow from requirement localization through cross-artifact editing. The best agent achieves 22.0% Pass@1 with guidance but only 7.3% without, indicating that both workflow discovery and ECAD execution remain substantial challenges for current agents. Beyond PCB, the workflow-grounded, expert-constructed evaluation approach may serve as a reference for other engineering domains where historical traces are sparse and correctness spans heterogeneous artifacts. Data and code are available at https://anonymous.4open.science/r/pcbinnobench-C841 .

Existing code benchmarks measure whether an agent can produce any test that reproduces a known bug, or whether it can produce a patch that fixes a described issue. Neither isolates the distinct skill of property-based testing: deriving a semantic invariant from documentation, and then constructing an input-generation strategy precise enough to make a random search reveal the violation. We introduce PBT-Bench, a benchmark of 100 curated property-based testing problems across 40 real Python libraries. Each problem injects one or more semantic bugs (365 in total, mean 3.65 per problem) designed so that default-strategy random inputs almost never trigger them; the agent must read the library's documentation, identify the relevant invariant, and specify a Hypothesis @given strategy that concentrates mass in the trigger region. Bugs are stratified across three difficulty levels (L1–L3) spanning single-constraint boundary bugs to stateful, cross-function protocol violations. We evaluate eight contemporary LLMs under two prompting regimes (open-ended baseline vs. explicit Hypothesis scaffolding) for three independent runs per configuration. Bug recall under the PBT-guided prompt ranges from 42.1% to 83.4% across models; under the open-ended baseline, from 31.4% to 76.7%. Hypothesis scaffolding lifts mid-capability models by over 20 percentage points, but yields smaller gains for the strongest models, with two exceptions showing degradation, suggesting the structured prompt can interfere with certain model behaviours rather than complementing them. The hardest bugs prove model-specific: different architectures fail on different problems, leaving persistent gaps that no single model closes. We release the benchmark, harness, and full evaluation corpus to support downstream work on documentation-grounded semantic reasoning.

A significant hurdle for current LLMs is the execution of complex, multi-stage tasks. Group Relative Policy Optimization (GRPO) has been emerging as a leading choice, but its reliance on sparse outcome rewards severely limits credit assignment across intermediate steps. Existing remedies—running full rollouts to assign step-level advantages, calling external LLM judges at each step, or computing intrinsic rewards that require ground-truth answers at every evaluation—introduce significant costs or practical constraints. We hypothesize that internal correctness probing over LLM hidden states can be repurposed as a step-level reward signal, potentially addressing all of these limitations at once. However, existing probing research assumes clean inputs, and we first show that this assumption breaks down in multi-step settings: hidden-state probes degrade severely under prefix contamination—tracking coherence with the (possibly corrupted) prefix rather than grounded correctness—while attention-based features remain robust to contamination but underperform on clean prefixes. Building on this complementary relationship, we propose the Prefix-Aware Internal Reward (PAIR)—a two-stage model with a frozen hidden-state probe estimating belief-consistency and a lightweight attention-based head correcting it toward grounded correctness. Experimental results show that PAIR achieves the highest AUROC on contaminated trajectories while operating at negligible inference cost, enabling dense step-level reward signals for GRPO training without external model calls, ground-truth dependencies, or full-trajectory rollouts.

Verified code generation asks a large language model (LLM) to generate both an executable program and a machine-checkable proof that the program meets a formal specification, promising software that is correct by construction. The de facto workflow decouples the two halves of the problem: first synthesize a program, then attempt to prove it correct. We observe that this sequential pipeline can be both ineffective and inefficient in practice. A program generated without anticipating its proof can be subtly incorrect or structurally difficult to verify, forcing the LLM into brittle repair loops that alternate between patching the code and patching the proof. Inspired by Dijkstra's view that a program and its correctness argument should be developed hand in hand, we propose \framework{}, an LLM-based agentic workflow that first derives a unified program-and-proof plan from the specification, then elaborates the implementation and proof scaffold under this shared plan. To evaluate verified code generation in realistic settings, we further introduce Lean4Commit0, a project-level benchmark built by extracting core APIs from real-world software repositories and translating their requirements, including relational specifications across APIs, into Lean tasks. Using four frontier LLM backends, we evaluate \framework{} on Verina, AlgoVeri, and our Lean4Commit0 benchmark, where it achieves the highest solve rate in every benchmark--model setting. Compared with the stronger baseline, it improves solve rates by 4.6--11.2 percentage points and reduces per-task API cost by up to roughly 40\% and wall-clock time by up to roughly 37\% on the difficult subset of each benchmark.

Retrieval-augmented generation (RAG) is increasingly used to ground Large language Models (LLMs) with up-to-date or private information, and has become a key component for knowledge-intensive LLM systems (e.g., LLM agents). However, RAG with LLMs often hallucinate due to unreliable retrieved information or outdated internal knowledge, making hallucination detection crucial for reliable LLM workflows. Recently, reasoning-intensive LLMs that produce explicit reasoning trajectories before answering have shown promise in RAG by examining and synthesizing retrieved evidence. Yet, they often suffer from overthinking: generating redundant or unproductive reasoning with limited performance gains. While prior work mainly studies overthinking in closed-book reasoning tasks that mainly rely on internal knowledge, overthinking in RAG-based question answering (QA) that requires open-book reasoning over both internal knowledge and external context remains underexplored. Thereby, we conduct empirical studies to investigate the relationship between overthinking and hallucination in RAG-based QA and find: (1) Overthinking in reasoning LLMs produces lengthy reasoning trajectories that are correlated with hallucinations; (2) This overthinking is often associated with knowledge conflict, where LLMs struggle to reconcile retrieved context with internal knowledge. Based on our findings, we propose a knowledge conflict-aware hallucination detection framework. We train a classifier to identify the three knowledge conflict reasoning patterns, design a knowledge conflict score to measure knowledge conflict severity, and apply tailored detection strategies with different prompts for different conflict levels. Unlike existing methods, our approach requires neither LLM parameter access nor retrieval reliability, and only needs the LLM output in a single run. Experiments on three QA datasets and three LLMs confirm our framework's effectiveness.

Scaling multi-agent systems driven by large language models is transitioning from lossy text-based communication to continuous latent interaction. However, existing latent interaction approaches relying on concatenation-based Softmax cross-attention suffer from an Attention Distraction Dilemma: due to the inherent anisotropy of latent spaces and the semantic misalignment induced by agent role heterogeneity, non-linear Softmax structurally amplifies irrelevant tokens, diluting the collaborative signal as interactions deepen. To address this issue, we propose a training-free additive mechanism that models upstream representations as a semantic manifold. By shifting from concatenation-based Softmax competition to linear subspace projection, this perspective inherently bypasses scalar noise accumulation. Specifically, we introduce two components: a Resonance Filter that extracts principal semantic directional alignments to overcome latent anisotropy, and an SNR-Adaptive Spectral Gain that dynamically regulates this projection to counteract role heterogeneity. We analyze this approach using the Asymptotic SNR Tipping Point Theorem, illustrating the scaling advantages of the mechanism beyond a critical noise threshold. Experiments indicate that the proposed approach outperforms baselines in 46 of 54 evaluated configurations, maintaining semantic reasoning in long-context collaborations.

Multi-agent LLM systems built on frontier backbones often fail to improve reliably as the swarm grows: under strong backbone models, performance frequently plateaus or declines beyond a small number of agents. In a budget-matched controlled intervention on SWE-bench Verified ($N{=}8$, Claude Opus 4.5 backbone, 64-call budget per instance), varying only the communication policy moves resolve rate from 80.4\% under random broadcast coordination (below the 80.9\% single-shot baseline) to 85.5\% under a learned orchestrator, a 5.1-point spread under matched compute. We introduce OrchestraRL, an orchestration layer trained over frozen LLMs. A centralized Macro-Orchestrator reads the swarm's context entropy $H(t)$, computed from the spread of agent output embeddings, and outputs communication mode (via learned entropy thresholds) and topology; per-agent Micro-Coordinators decide whom each agent addresses, when to speak, and at what granularity. Both are trained with REINFORCE and a learned value baseline. Across four benchmarks (SWE-bench Verified, GPQA Diamond, LiveCodeBench v6, SimpleQA), OrchestraRL is comparable to the strongest baseline at $N{=}2$ and achieves the best resolve rate at $N \in \{4, 8\}$ on every benchmark, with the gap widening as $N$ grows. The learned policy recovers interpretable task-dependent communication patterns, including clustered exploration, star-based synthesis, and pipeline reasoning, without these structures being hand-coded.

We demonstrate that AI agents can autonomously improve mathematical proofs of quantitative results. As a key step towards this frontier, we achieve strict improvements on Grothendieck's and Korenblum's constants, whose precise values have remained surprisingly elusive despite their recurrence in both the mathematical and physical sciences. By encoding relevant literature as a partially formalized Lean blueprint, we structure the agent's search space to localized proof refinements that automatically propagate to a verified final bound. Concretely, Grothendieck's constant ($\mathcal{K}_G$) is shown to be: $$ \mathcal{K}_G \leq \tfrac{\pi}{2 \log (1 + \sqrt{2})} - 10^{-17}, $$ which is the first explicit numerical improvement since Krivine's landmark 1979 result. For Korenblum's constant ($\mathcal{K}_K$), the lower bound is tightened to: $$ \mathcal{K}_K \geq 0.3554 + 0.001165, $$ a third-digit improvement over Wang's recent 2025 result. In summary, we establish that AI agents can autonomously refine formal proofs of open problems and provide a generalizable framework for future mathematical discovery.

Tool documentation is the critical interface bridging Large Language Model (LLM) agents and external environments. However, real-world documentation is often ambiguous or insufficient, leading to misaligned tool calls and task failures. While existing methods rely on parameter-efficient tuning or reinforcement learning to enhance tool-use capabilities, they incur substantial computational overhead and lack the agility to adapt to frequently updated toolkits. We propose OpTool , an evolutionary framework that treats tool documentation as an evolvable configuration. OpTool consists of three stages: (1) Contrastive Trajectory Generation, which explores diverse tool-use trajectories via contrastive beam search; (2) Trajectory-based Insight Extraction, which extracts tool-use insights through retrospective global credit assignment; and (3) Tool Documentation Evolution, which iteratively refines the documentation to mitigate information insufficiency. Experimental results across multiple domains demonstrate that OpTool significantly improves tool call accuracy and task success rates, providing a robust solution for tool-based LLM agents.

Large Language Model-based Multi-Agent Systems (LLM-based MAS) have shown remarkable success in solving complex tasks by coordinating specialized agents through multi-step workflows. To overcome the high cost and specificity of manually designed workflows, recent research has shifted toward learning them automatically. However, a key limitation remains: while agent calls introduce significant latency due to model inference time and service congestion, most automated methods optimize solely for downstream performance, ignoring execution time (makespan). This issue is amplified when workflows are reused across recurring queries, making latency accumulate over time. This limits deployment in recurrent applications such as AI customer support and AI cloud troubleshooting, where users expect fast responses. To bridge this gap, we introduce a new problem, termed latency-aware workflow generation, where the objective is to minimize workflow makespan while aiming to retain downstream performance. We propose LAWA (Latency-Aware Workflow optimizAtion), a novel framework that reduces makespan via theoretically grounded graph edits targeting critical-path bottlenecks. By unifying a graph generator with this graph-editing procedure, LAWA supports both de novo workflow generation and refinement of existing ones. Empirical results across seven benchmarks demonstrate that LAWA significantly reduces makespan while achieving competitive task performance, supporting efficient reuse of multi-agent workflows for recurrent queries.

We study contextual dynamic pricing with linear valuations and bounded-support agnostic noise, whose induced demand curve may be non-Lipschitz with arbitrary jumps and atoms. Such discontinuities break the cross-context interpolation arguments used by smooth-demand pricing algorithms, while the best previous method achieved only $\widetilde O(T^{3/4})$ regret. We propose Conservative-Markdown Redirect-UCB Pricing, a polynomial-time algorithm that combines randomized parameter estimation, conservative residual-grid probing, and confidence-based one-step redirection. Our algorithm achieves $\widetilde O(T^{2/3})$ optimal regret, matching the known lower bounds of Kleinberg and Leighton (2003) up to logarithmic factors and improving over the previous upper bound of Xu and Wang (2022). Under stochastic well-conditioned contexts, this closes the long-existing open regret gap in linear-valuation contextual pricing under agnostic non-Lipschitz noise distribution.

Deep search capabilities have become an indispensable competency for frontier large language model (LLM) agents, yet training high-performance deep search agents remains constrained by three key challenges: as synthesized reasoning graphs grow in complexity, verifying answer correctness and uniqueness becomes increasingly difficult to scale; heavy dependence on online search APIs makes data synthesis costly and hard to scale; and the lack of fully open synthesis pipelines severely hinders reproducibility.To bridge this gap, we introduce OpenSearcher, the first deep search agent training pipeline with zero online API dependency throughout the entire data synthesis and model training process, built upon two core technical innovations: (1) \textit{Distinguishing Set-grounded QA synthesis with programmatic verification}: we compute minimal Distinguishing Sets on Wikidata to uniquely identify target entities, recursively expand them into reasoning DAGs, and transform the structured graphs into natural language questions via progressive generation and fuzzification. Answer uniqueness is deterministically verified through SPARQL conjunctive queries, complemented by lightweight LLM filtering for linguistic quality. (2) \textit{Fully offline training pipeline}, which builds an offline corpus of tens of millions of documents from publicly downloadable Wikidata, Wikipedia, and Web Crawl Dumps, and completes trajectory synthesis, SFT cold start, and reinforcement learning entirely offline. Experimental results demonstrate that OpenSearcher achieves strong performance across deep search benchmarks, establishing a new state of the art among fully offline methods, for example, achieving 32.3\% on BrowseComp and 72.8\% on GAIA. We fully open-source the complete data synthesis pipeline to democratize offline deep search agent research and foster a more transparent, collaborative ecosystem.

Deep search has become a crucial capability for frontier multimodal agents, enabling models to solve complex questions through active search, evidence verification, and multi-step reasoning. Despite rapid progress, top-tier multimodal search agents remain difficult to reproduce, largely due to the absence of open high-quality training data, transparent trajectory synthesis pipelines, or detailed training recipes. To this end, we introduce **OpenSearch-VL**, a fully open-source recipe for training frontier multimodal deep search agents with agentic reinforcement learning. First, we curate a dedicated pipeline to construct high-quality training data through Wikipedia path sampling, fuzzy entity rewriting, and source-anchor visual grounding, which jointly reduce shortcuts and one-step retrieval collapse. Based on this pipeline, we curate two training datasets, *SearchVL-SFT-36k* for SFT and *SearchVL-RL-8k* for RL. Besides, we design a diverse tool environment that unifies text search, image search, OCR, cropping, sharpening, super-resolution, and perspective correction, enabling agents to combine active perception with external knowledge acquisition. Finally, we propose a multi-turn fatal-aware GRPO training algorithm that handles cascading tool failures by masking post-failure tokens while preserving useful pre-failure reasoning through one-sided advantage clamping. Built on this recipe, OpenSearch-VL delivers substantial performance gains, with over 10-point average improvements across seven benchmarks, and achieves results comparable to proprietary commercial models on several tasks. We will release all data, code, and models to support open research on multimodal deep search agents.

Scientific discovery is fundamentally open-ended, characterized by epistemic uncertainty, and requiring endogenous goal-setting and recursive knowledge accumulation. While current agentic systems have demonstrated success in addressing goal-based discovery tasks, they struggle to sustain discovery in open-ended settings. In this work, we posit that open-ended discovery is an emergent network-level property of a socially driven system. We introduce $\texttt{ASCollab}$, which translates four key algorithmic conditions: (1) population heterogeneity in research behaviors, (2) endogenous interaction networks where collaborations and attention routing emerge organically, and (3) socially driven peer evaluations, which are enabled by (4) a global, shared memory that reflects evolving research states. Specifically, social memory is implemented to capture time-varying expertise and reputational signals, and shifting attention paid to historical artifacts produced by the system. Through experiments on large-scale discovery problems in genomics, cell biology, and epidemiology, we demonstrate that $\texttt{ASCollab}$ produces discoveries judged by domain experts to be both sound and significant. Crucially, the system independently recovers real-world scientific findings. Furthermore, systematic investigations show that social dynamics fostered by each of the four conditions are vital, where heterogeneous agents operating within self-organizing networks significantly outperform both fixed-workflow baselines and homogeneous networks.

Large language model (LLM)-based robot task planning is promising for open-ended instruction following, but degrades on long-horizon tasks in large environments. When spatial information is conveyed to the LLM through text, the model can fail to capture spatial context, and token cost grows with environment size. Generating action sequences directly with an LLM also makes it difficult to satisfy the current world state and action preconditions. We address this with an ontology-grounded scene representation that aligns objects, spaces, relations, and states in a shared symbolic vocabulary for spatial reasoning and task planning, and with OntoPlan, an agentic framework that interprets instructions, selectively retrieves task-relevant information, formalizes goals and constraints, and produces executable plans. Across 150 benchmark tasks spanning five indoor environments and three scene scales, OntoPlan achieves 0.83 average task success, compared with 0.30 for the strongest baseline, while using 17.2k total tokens per task on average, about 5.8$\times$ fewer than the most efficient baseline. These advantages persist as scene scale increases, whereas prior methods degrade more sharply in both success and token cost. OntoPlan also responds appropriately to ambiguous or infeasible instructions by asking follow-up questions or reporting insufficient information rather than committing to invalid plans. Code available at https://anonymous.4open.science/r/OntoPlan.