ScholarPulse 日报 2026-11-10

2026-11-10 学术简报：2 篇。论文提出评测基准 WorkflowArena，聚焦多智能体系统中大模型自主编排可执行工作流的能力，证实工作流组织本身是决定任务性能的关键因素。

今日速览

序号	标题	来源	日期	主题	推荐等级
1	WorkflowArena: Benchmarking Executable Workflow Orchestration in LLM-Based Multi-Agent Systems	arXiv	2026-11-09	AI-Agent	高
2	Agentic-OPD: Recovering Forgotten Agent Capabilities via On-Policy Distillation for Agentic Tasks	arXiv	2026-11-09	RAG	高

重点论文与技术动态

1. WorkflowArena: Benchmarking Executable Workflow Orchestration in LLM-Based Multi-Agent Systems

来源：arXiv
日期：2026-11-09
作者/机构：Zheng Jiang, Fang Niu, Bo Peng, Houde Qian, Wenyi Li, Deyao Hong
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.CL, cs.SE

一句话结论

论文提出评测基准 WorkflowArena，聚焦多智能体系统中大模型自主编排可执行工作流的能力，证实工作流组织本身是决定任务性能的关键因素。

核心内容

形式化“工作流编排”能力，评估模型在运行时自主派生子智能体并编写协调流程的水平。
评测 9 款模型发现不存在全领域最优配置，且策略差异显著（如 Claude 偏好大组低频委托，GPT 偏好小组高频委托）。
强模型（Claude Opus 5.5）编写的工作流能显著提升弱执行者的任务表现，证明组织架构具有独立的度量价值。

方法与数据

涵盖软件工程、形式数学等 8 个专业领域的 145 个长周期任务；依托隐藏验证器评估复合交付物，并全程记录委托行为以供联合分析。

价值判断

值得关注：将多智能体评测由预设固定拓扑推进到模型自主动态生成可执行工作流的新维度。
可复用点：结合成果质量与组织过程记录的双重评测范式。
局限/待核查：隐藏验证器的具体实现细节及更多模型泛化性在摘要中未明确。

摘要

Large language model agents are increasingly deployed as multi-agent systems rather than as a standalone model operating within a single context. Recent agent harnesses allow the model to instantiate worker agents and to author the executable workflow that coordinates them, shifting the design of system structure from developers to the model at run time. The quality of this organization therefore constitutes a distinct capability, yet existing benchmarks either measure end-to-end task completion or evaluate coordination under structures specified in advance. We formalize this capability as workflow orchestration and introduce WorkflowArena, a benchmark of 145 long-horizon tasks spanning 8 professional domains, from software engineering and formal mathematics to legal analysis and 3D production. Each task specifies an initial workspace and an integrated deliverable composed of interdependent components, which is assessed by a hidden verifier. Agents determine autonomously whether and how to delegate, and all delegation is recorded, enabling joint evaluation of deliverable quality and the organization of the underlying work. Experiments with nine models show that no single configuration performs best across all domains and that systems exhibit distinct delegation strategies: at high effort, Claude models in Claude Code invoke workflows less frequently but with larger worker groups, whereas GPT models in Codex delegate more frequently in smaller units. When the executing model is held fixed on a challenging subset of tasks, workflows authored by Claude Opus 5.5 substantially improve weaker executors, indicating that workflow organization is itself a measurable determinant of task performance.

2. Agentic-OPD: Recovering Forgotten Agent Capabilities via On-Policy Distillation for Agentic Tasks

来源：arXiv
日期：2026-11-09
作者/机构：Tingcheng Bian, Kangye Gong, Jinchang Luo, MingQuan Cheng, Feiran Liu, Ni Li
主题标签：RAG, arXiv
推荐等级：高
分类：cs.LG, cs.AI

一句话结论

Agentic-OPD 利用强化学习中废弃的历史检查点作为零成本教师池，通过在线策略蒸馏有效恢复智能体遗忘的解题能力。

核心内容

发现 RL 训练中各检查点解题能力的并集远超单一最佳模型，后者常遗忘早期检查点已能解决的问题。
提出 HCN 策略，依据对学生未解问题的加权覆盖度动态挑选互补的教师检查点。
引入 LAD 模块，仅在教师更优且学生失败的任务上蒸馏，防止覆盖已掌握的技能。

方法与数据

以最佳验证模型为学生、历史检查点为冻结教师进行在线策略蒸馏；在 5 个任务领域的 24 个基准与子集上评测，最高提升 7.4 分，4B 学生模型在深度搜索上超越了 GPT-5.6 和 DeepSeek-V4.1。

价值判断

值得关注：确立了与强化学习算法本身正交的“后 RL 能力恢复”新阶段，零额外训练成本挖掘历史检查点价值。
可复用点：基于未解问题覆盖的教师选择（HCN）与局部感知蒸馏（LAD）机制。
局限/待核查：摘要对比了 GPT-5.6 和 DeepSeek-V4.1 等前瞻模型，具体的评测基准与实验设定需核查正文。

摘要

Reinforcement learning is now the mainstream paradigm for training LLMs on agentic tasks, yet standard practice keeps only the best-validation checkpoint and discards the rest. We show that this is lossy: in GRPO training of LLM agents, the best-validation checkpoint fails on problems that earlier checkpoints from the same run had already solved, and the union of problems solved across checkpoints far exceeds any single one. The discarded checkpoints thus form a teacher pool obtained at no additional training cost, which on-policy distillation (OPD) can mine to recover the lost coverage. Mining this pool raises two challenges: i) which checkpoint to distill from, as no checkpoint dominates the others and complementarity to the student is problem-specific; ii) where to trust it, as even a well-chosen teacher is only locally superior, so indiscriminate distillation overwrites behavior the student has already mastered. We propose Agentic-OPD, which treats the best-validation checkpoint as the student and all saved checkpoints as frozen teacher candidates, using the student's evolving capability gaps to guide both teacher selection and supervision filtering: the HCN policy (Highest Coverage Next) dynamically selects teachers by weighted coverage of the student's unsolved problems, while the LAD module (Locality-Aware Distillation) confines transfer to failed student rollouts on problems where the teacher performs better. Across five agentic task domains and 24 benchmarks and task subsets, Agentic-OPD improves over its initialization checkpoint by up to 7.4 points, with a 4B student surpassing the latest frontier models GPT-5.6 and DeepSeek-V4.1 on deep search. Agentic-OPD recasts a finished RL run as a reusable pool of policies and establishes post-RL capability recovery as a stage orthogonal to improvements in the RL algorithm itself.