LongCat-DeepResearch Technical Report
打开互动全文版（逐段中英对照 + 图/公式 + 针对本篇提问）→我们提出 LongCat-DeepResearch，一个将增强版 LongCat 模型与多智能体工作流相结合的深度研究系统，用于生成全面且以证据为基础的报告。该工作流将全局规划与细节调研分离，并在章节层面协调修订。多个规划智能体首先探索外部来源，并细化出一份可执行的研究计划，称为 ResearchSpec。随后，研究智能体并行调研并起草各自负责的章节，并在分析推进过程中于独立上下文中收集更多证据。各章节汇总后，全局审阅会指导有针对性的局部修订，从而减少对整篇报告反复重写的依赖。该工作流还支持为 LongCat 通用模型的中期训练与后期训练构建研究任务和轨迹。LongCat-DeepResearch 在 DeepResearchBench 上取得 55.25 分，在 DeepResearchBench II 上取得 51.35 分，在 ResearchRubrics 上取得 79.83 分。在一项内部基准上，它获得 76.04 分，在四个对比系统中排名第二。开发集分析表明，融合多种规划视角能带来收益，而进一步细化规划的效果则好坏参半。额外的编辑处理提升了两个基准上的平均自动可读性偏好，但在两个基准上的趋势有所不同。
We present LongCat-DeepResearch, a deep research system that combines an enhanced LongCat model with a multi-agent workflow for producing comprehensive, evidence-grounded reports. The workflow separates global planning from detailed investigation and coordinates revision at the section level. Multiple planning agents first explore external sources and refine an actionable research plan, termed ResearchSpec. Research agents then investigate and draft their assigned sections in parallel, gathering additional evidence in separate contexts as their analyses develop. Once the sections are assembled, global review guides targeted local revisions, reducing reliance on repeated full-report rewriting. This workflow also supports the construction of research tasks and trajectories for the mid-training and post-training of LongCat's general-purpose models. LongCat-DeepResearch achieves 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics. On an in-house benchmark, it scores 76.04, ranking second among four compared systems. Development-set analyses show benefits from combining planning perspectives, while further planning refinement has mixed effects. Additional editing improves average automatic readability preference across two benchmarks, with different trends on each.
我们提出 LongCat-DeepResearch，一个深度研究系统，它将增强的 LongCat 模型与多智能体工作流相结合，用于生成全面、有证据支撑的报告。该工作流将全局规划与详细调查分离，并在章节层面协调修订。多个规划智能体首先探索外部来源，并细化出一份可执行的研究计划，称为 ResearchSpec。随后，研究智能体并行地调查并起草各自负责的章节，随着分析推进，在各自独立的上下文中收集额外证据。各章节汇总后，全局审阅指导有针对性的局部修订，从而减少对反复重写整份报告的依赖。该工作流还支持为 LongCat 通用模型的中间训练和后训练构建研究任务与轨迹。LongCat-DeepResearch 在 DeepResearchBench 上达到 55.25，在 DeepResearchBench II 上达到 51.35，在 ResearchRubrics 上达到 79.83。在一个内部基准上，它得分为 76.04，在四个对比系统中排名第二。开发集分析表明，结合多种规划视角带来收益，而进一步的规划细化则效果不一。额外的编辑提升了两个基准上的平均自动可读性偏好，但在各基准上趋势不同。
We present LongCat-DeepResearch, a deep research system that combines an enhanced LongCat model with a multi-agent workflow for producing comprehensive, evidence-grounded reports. The workflow separates global planning from detailed investigation and coordinates revision at the section level. Multiple planning agents first explore external sources and refine an actionable research plan, termed ResearchSpec. Research agents then investigate and draft their assigned sections in parallel, gathering additional evidence in separate contexts as their analyses develop. Once the sections are assembled, global review guides targeted local revisions, reducing reliance on repeated full-report rewriting. This workflow also supports the construction of research tasks and trajectories for the mid-training and post-training of LongCat's general-purpose models. LongCat-DeepResearch achieves 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics. On an in-house benchmark, it scores 76.04, ranking second among four compared systems. Development-set analyses show benefits from combining planning perspectives, while further planning refinement has mixed effects. Additional editing improves average automatic readability preference across two benchmarks, with different trends on each.
开放式深度研究要求语言模型智能体探究界定不清的问题，并从多样化的外部来源收集证据。这些智能体将其发现综合为全面的长篇报告。与传统的信息检索不同，此类报告的证据需求无法在调查开始前完全确定。随着分析的推进，新出现的解释会暴露缺失的证据，未解决的论断需要进一步验证，新的研究问题也会随之产生。因此，研究系统必须建立一个有用的初始议程，同时允许在各章节展开过程中继续收集证据。这引出一个实际问题：深度研究智能体如何在协调共享研究议程的同时，保留撰写每个章节所需的细致调查？
Open-ended deep research requires language-model agents to investigate underspecified questions and gather evidence from diverse external sources. These agents synthesize their findings into comprehensive long-form reports. Unlike conventional information seeking, the evidence requirements of such reports cannot be fully specified before the investigation begins. As an analysis develops, emerging explanations expose missing evidence, unresolved claims require further verification, and new research questions arise. A research system must therefore establish a useful initial agenda while allowing further evidence gathering as its sections develop. This raises a practical question: How can deep-research agents coordinate a shared research agenda while preserving the detailed investigation needed to write each section?
现有系统以不同方式实现研究与写作之间的这种反馈。先前的方法将推理、检索和起草交错进行，或利用不断演化的报告来指导后续的检索与修订。其他方法则在详细撰写之前，通过多视角和基于证据的提纲来组织调查。当这种反馈依赖于不断增长的研究历史或反复更新完整报告时，会出现三个困难。第一，证据、中间推理和草稿文本争夺有限的上下文窗口；压缩可能移除那些相关性尚未显现的信息。第二，早期发现可能锚定正在发展的叙事并引导后续搜索，从而缩小所探索的视角范围。第三，反复更新完整报告会重新生成新证据所影响段落之外的文本。挑战在于：既要保留研究与写作之间的反馈，又不将整个调查集中于一份不断膨胀的草稿中。
Existing systems implement this feedback between research and writing in different ways. Prior approaches interleave reasoning, retrieval, and drafting, or use an evolving report to guide subsequent retrieval and revision. Other approaches organize investigation through multiple perspectives and evidence-grounded outlines before detailed composition. When this feedback relies on a growing research history or repeated full-report updates, three difficulties arise. First, evidence, intermediate reasoning, and draft text compete for finite context; compression may remove information whose relevance has not yet become apparent. Second, early findings can anchor the developing narrative and steer subsequent searches, narrowing the perspectives explored. Third, updating the full report repeatedly regenerates text beyond the passages affected by new evidence. The challenge is to retain feedback between research and writing without concentrating the entire investigation in an expanding draft.
在本工作中，我们提出 LongCat-DeepResearch，一个将改进的 LongCat 模型与以可执行 ResearchSpec 为中心的研究框架相结合的系统。其核心思想是将早期迭代从完整报告转移到对其研究需求的紧凑规格说明上。多个规划器独立搜索并阅读外部来源，以形成候选规格。它们的提案经过整合与精炼，以识别缺失的问题、证据需求和章节职责。这一过程将任务中正在形成的全局理解编码到 ResearchSpec 中，后者作为报告所需确立内容的紧凑代理。下游综合流水线通过组装与编辑保留完整的章节草稿。每位研究员接收完整的 ResearchSpec 和一项任务分配，在独立的上下文中继续收集证据，并根据其发现撰写一个完整章节。由此产生的章节产物被组装成草稿，随后由 Global Editor 识别跨章节问题，并由 Local Editors 进行针对性修订。这一设计将紧凑的全局协调与细致的局部研究相结合。ResearchSpec 在章节派发前得到精炼；后续调查在分配的章节内进行，编辑则协调其文本，而无需重新开启全局研究议程。
In this work, we introduce LongCat-DeepResearch, a system combining an improved LongCat model with a research harness centered on an executable ResearchSpec. The key idea is to shift early iteration from the full report to a compact specification of its research requirements. Multiple planners independently search and read external sources to develop candidate specifications. Their proposals are consolidated and refined to identify missing questions, evidence requirements, and section responsibilities. This process encodes the emerging global understanding of the task into ResearchSpec, which serves as a compact proxy for what the report needs to establish. The downstream synthesis pipeline retains complete section drafts through assembly and editing. Each researcher receives the complete ResearchSpec and one assignment, continues gathering evidence in an independent context, and writes a complete section from its findings. The resulting section artifacts are assembled into a draft, after which a Global Editor identifies cross-section issues and Local Editors perform targeted revisions. This design combines compact global coordination with detailed local research. ResearchSpec is refined before section dispatch; subsequent investigation proceeds within the assigned sections, and editing coordinates their text without reopening the global research agenda.
我们在三个既有基准和一个内部基准上评估 LongCat-DeepResearch。它在 DeepResearchBench 上达到 55.25，在 DeepResearchBench II 上达到 51.35，在 ResearchRubrics 上达到 79.83，在记录的系统级比较中（第 4.3 节）分别超过三个对比深度研究产品中最强者 0.30、3.17 和 5.62 分。在我们的内部基准上，LongCat-DeepResearch 以 76.04 的总分排名第二，超过 Claude-DeepResearch 的 61.42 和 Gemini-DeepResearch 的 42.49，落后 ChatGPT-DeepResearch 的 76.59 仅 0.55 分。相同的阶段接口还支持构建用于 LongCat 通用模型中期训练和后训练的研究问题、任务特定评分标准和轨迹。
We evaluate LongCat-DeepResearch on three established benchmarks and an in-house benchmark. It achieves 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics, outperforming the strongest of the three compared deep-research products by 0.30, 3.17, and 5.62 points, respectively, in the recorded system-level comparison (Section 4.3). On our in-house benchmark, LongCat-DeepResearch ranks second with an overall score of 76.04, exceeding Claude-DeepResearch at 61.42 and Gemini-DeepResearch at 42.49, and trailing ChatGPT-DeepResearch's 76.59 by 0.55 points. The same stage interfaces also support the construction of research questions, task-specific rubrics, and trajectories used in the mid-training and post-training of LongCat's general-purpose models.
长上下文访问并不能确保在整个输入中可靠地使用证据。MemGPT 将工作上下文与外部存储分离，而 AgentFold 学习细粒度的上下文折叠，RE-TRAC 在研究尝试中携带结构化摘要。FoldAct 研究了当生成的摘要改变未来观察时引入的训练不稳定性。AdaCoM 进一步表明，优选的压缩程度随底层智能体而变化。这些工作使压缩成为一种显式、有用的操作，而非附带的实现细节。在深度研究中，SearchSwarm 向编排器返回紧凑的、基于引用的子智能体报告，Argus 从紧凑的证据图合成答案。Deep-Reporter 在顺序生成章节时维护循环的全局摘要和逐字局部尾部。我们的报告合成路径在组装和全局到局部编辑过程中保留完整的章节产物，而外部记忆和压缩则处理研究历史的单独管理。
Long-context access does not ensure reliable use of evidence throughout the input. MemGPT separates working context from external storage, while AgentFold learns fine-grained context folding and RE-TRAC carries structured summaries across research attempts. FoldAct studies the training instability introduced when generated summaries change future observations. AdaCoM further shows that the preferred degree of compression varies with the underlying agent. These works make compression an explicit, useful operation rather than an incidental implementation detail. In deep research, SearchSwarm returns compact, citation-grounded subagent reports to an orchestrator, and Argus synthesizes an answer from a compact evidence graph. Deep-Reporter maintains a recurrent global summary and a verbatim local tail during sequential section generation. Our report-synthesis path retains complete section artifacts through assembly and global-to-local editing, while external memory and compression address the separate management of research history.
WebGPT 和 ReAct 建立了基于工具的交互。STORM 在文章写作前发展观点，MindSearch 通过可执行图分解搜索。OmniThink 使用信息树和概念池将知识扩展与反思连接起来。最近的工作使研究结构可变：WebWeaver 将证据获取与大纲优化交错进行，AgentCPM-Report 交替起草和深化，ScaffoldAgent 使用下游效用选择大纲更改。Enterprise Deep Research 结合了覆盖目标、依赖引导的信息共享和证据充分性条件；SearchOS 外化覆盖、证据、待处理任务和失败的搜索。DecomposeR 使类型化研究 DAG 可通过结构感知奖励进行训练。
WebGPT and ReAct established tool-grounded interaction. STORM develops perspectives before article writing, and MindSearch decomposes search through an executable graph. OmniThink uses an information tree and a conceptual pool to connect knowledge expansion with reflection. More recent work makes the research structure mutable: WebWeaver interleaves evidence acquisition with outline optimization, AgentCPM-Report alternates drafting and deepening, and ScaffoldAgent selects outline changes using downstream utility. Enterprise Deep Research combines coverage objectives, dependency-guided information sharing, and evidence-sufficiency conditions; SearchOS externalizes coverage, evidence, pending tasks, and failed searches. DecomposeR makes a typed research DAG trainable with structure-aware rewards.
最接近的结构化状态比较还包括 RhinoInsight，其可验证的检查表约束研究行动，其证据审计将来源与起草内容联系起来，以及 DualGraph，它维护独立、共同进化的知识和大纲图。这些系统将结构化规划和证据绑定确立为现有的设计选择。在我们的流程中，ResearchSpec 定义了规划后固定的章节级执行职责；研究人员将他们的发现保留在传递给编辑协调的章节产物中。其角色是执行接口，而不是持久知识图或用于个性化查询细化的输入级规范。
The closest structured-state comparisons also include RhinoInsight, whose verifiable checklists constrain research actions and whose evidence audit links sources to drafted content, and DualGraph, which maintains separate, co-evolving knowledge and outline graphs. These systems establish structured planning and evidence binding as existing design choices. In our pipeline, ResearchSpec defines section-level execution responsibilities that are fixed after planning; researchers retain their findings in section artifacts passed to editorial reconciliation. Its role is an execution interface, rather than a persistent knowledge graph or an input-level specification for personalized query refinement.
PaperQA2 开发了引用的科学主题摘要和文献矛盾检测；OpenScholar 结合了文献检索、引用支持的生成和迭代自我反馈。这些系统将科学综合确立为超越孤立事实检索的研究目标。在开放式报告写作中，WebThinker 将推理、搜索和起草交错进行，而 TTD-DR 使用不断演变的草稿来指导检索和迭代报告修订。FS-Researcher 将持久证据收集与多会话报告写作分开，并保留原始源文件。CogGen 协调全局计划-写作-审查循环与章节级工作，并允许全局重构；Ptah 维护可检查的研究产物和用于多模态组合的视觉记忆。我们的设计在决定如何协调之前组装独立编写的章节：全局编辑器指定所有权，而局部编辑器修订分配的单位。此范围不同于在一次调用中重新生成整个文档，但不保证编辑保留每个有用的细节。Mr. Dre 提供了直接证据，表明报告修订可以满足新反馈，同时损害早期覆盖或引用质量。DeepTRACE 审计语句级支持和归属，区分列出的来源与支持的声明。Self-Refine、DuMate 和 AREX 提供互补的反馈和验证循环。
PaperQA2 develops cited scientific-topic summaries and literature-contradiction detection; OpenScholar combines literature retrieval, citation-backed generation, and iterative self-feedback. These systems establish scientific synthesis as a research objective beyond isolated fact retrieval. In open-ended report writing, WebThinker interleaves reasoning, search, and drafting, while TTD-DR uses an evolving draft to guide retrieval and iterative report revision. FS-Researcher separates persistent evidence collection from multi-session report writing and retains original source files. CogGen coordinates a global plan–write–review loop with section-level work and permits global restructuring; Ptah maintains inspectable research artifacts and visual memory for multimodal composition. Our design assembles the independently written sections before deciding how to reconcile them: the Global Editor specifies ownership, while Local Editors revise assigned units. This scope differs from regenerating the entire document in one invocation, but does not guarantee that edits retain every useful detail. Mr. Dre provides direct evidence that report revision can satisfy new feedback while damaging earlier coverage or citation quality. DeepTRACE audits statement-level support and attribution, distinguishing listed sources from supported claims. Self-Refine, DuMate, and AREX offer complementary feedback and verification loops.
开放实现也为研究编排提供了实践先例。NVIDIA AI-Q 结合结构化规划、并发研究人员和专用写作者来生成引用支持的报告。LangChain 的 Open Deep Research 支持可配置的模型和搜索工具，具有压缩发现和撰写最终报告的独立阶段。GPT Researcher 将规划、证据收集和报告合成分开，并支持递归探索。Hugging Face 的 Open Deep Research 使用基于代码的智能体，具有网页浏览和文档检查工具。我们基于这些已建立的工作流模式，专注于共享规划以及完整章节草稿的保存和修订。
Open implementations also provide practical precedents for research orchestration. NVIDIA AI-Q combines structured planning, concurrent researchers, and a dedicated writer to produce citation-backed reports. LangChain’s Open Deep Research supports configurable models and search tools, with separate stages for compressing findings and writing the final report. GPT Researcher separates planning, evidence collection, and report synthesis, and supports recursive exploration. Hugging Face’s Open Deep Research uses code-based agents with web-browsing and document-inspection tools. We build on these established workflow patterns, focusing on shared planning and the preservation and revision of complete section drafts.
OpenAI 的 Deep Research System Card 描述了在浏览任务上的强化学习，其中包括使用评分标准（rubrics）评分的开放式任务。DR Tulu 提出了一种开放的长篇研究训练方法，其中评分标准随策略和新获得的证据而演变。这些是不同的先例：基于评分标准的监督早于演化评分标准的表述。Tongyi DeepResearch 和 Step-DeepResearch 描述了跨中期训练和后训练的合成任务与智能体式训练。S1-DeepResearch 强调规划、证据整合和报告生成，超越了以搜索为中心的问答，而 Marco DeepResearch 强调在任务和轨迹构建中的验证。对于开放式报告，要求必须得到支持并与问题对齐。ResearchRubrics 贡献了专家编写的提示和标准，包括隐含要求和负面标准；DeepResearchBench II 从专家文章中提取原子标准。这些评估资源与合成训练数据不同。DeepRubric 构建证据树并联合合成查询和评分标准，而 Quest 使用评分标准树来构建客观和开放式研究任务。我们的数据讨论基于这些原则，并描述了面向研究型训练数据的文档来源、问题-评分标准对齐和轨迹接口。
OpenAI's Deep Research System Card describes reinforcement learning on browsing tasks that include open-ended tasks graded with rubrics. DR Tulu develops an open long-form research training approach in which rubrics evolve with the policy and newly acquired evidence. These are distinct precedents: rubric-based supervision predates the evolving-rubric formulation. Tongyi DeepResearch and Step-DeepResearch describe synthetic tasks and agentic training across mid-training and post-training. S1-DeepResearch emphasizes planning, evidence integration, and report generation beyond search-centric QA, while Marco DeepResearch emphasizes verification in task and trajectory construction. For open-ended reports, requirements must be supported and aligned with the question. ResearchRubrics contributes expert-written prompts and criteria, including implicit requirements and negative criteria; DeepResearchBench II derives atomic criteria from expert articles. These evaluation resources are distinct from synthetic training data. DeepRubric constructs evidence trees and jointly synthesizes queries and rubrics, whereas Quest uses rubric trees to construct both objective and open-ended research tasks. Our data discussion builds on these principles and describes document provenance, question–rubric alignment, and trajectory interfaces for research-oriented training data.
我们的方法在相互独立的研究上下文中协调细致的证据收集与报告构建。LongCat-DeepResearch 将各章节的研究分配到独立的上下文，并将全局编辑决策与局部文本生成分离。由此产生的中间产物也为数据构建和分阶段评估提供了单元。
Our method coordinates detailed evidence gathering and report construction across separate research contexts. LongCat-DeepResearch assigns section research to independent contexts and separates global editorial decisions from local text generation. The resulting intermediate artifacts also provide units for data construction and stage-wise evaluation.
ResearchSpec 在详细证据积累之前分配研究职责。独立研究者（Independent Researchers）撰写各自负责的章节，随后这些章节被汇编并编辑成一份报告（图 2）。保留这些章节产物并不意味着要保存每一个检索到的页面或底层交互的每一个细节。
ResearchSpec assigns research responsibilities before detailed evidence accumulates. Independent Researchers develop their assigned sections, which are then assembled and edited into a report (Fig. 2). Retaining these section artifacts does not imply preserving every retrieved page or every detail of the underlying interactions.
独立规划撰写者（规划者）在提出报告结构之前会简要搜索和阅读来源，将研究范围建立在模型先验知识之外的可用证据之上。他们各自的探索可以揭示互补的视角，正如在视角引导的文章规划中那样。规划评审（Planning Judge）整合候选规范，批评者（Critic）寻找遗漏的问题和案例，修订者（Reviser）纳入反馈。由于章节分配指导后续研究，这种细化旨在在某个方向被遗漏之前识别出疏漏。额外的候选方案和细化轮次提供了将更多计算分配给规划的方式；其收益并无保证。
Independent Planning Writers (planners) briefly search and read sources before proposing the report's structure, grounding the research scope in available evidence beyond the model's prior knowledge. Their separate explorations can reveal complementary perspectives, as in perspective-guided article planning. The Planning Judge consolidates the candidate specifications, the Critic searches for missing questions and cases, and the Reviser incorporates the feedback. Because section assignments direct subsequent research, this refinement seeks to identify omissions before a direction is left uninvestigated. Additional candidates and refinement rounds provide ways to allocate more computation to planning; their benefit is not guaranteed.
ResearchSpec 记录每个章节的范围、研究问题、所需实体或案例以及临时来源线索。这种对预期报告的紧凑表示还规定了研究工作如何跨上下文划分。其覆盖范围和职责可以在生成完整草稿之前被检查和修订。每个可调度的子章节都有一个唯一的层级 ID，例如 S1.2；我们将这些执行单元称为章节。在调度之前，验证器会检查 ID 唯一性、父子关系、连续顺序和必填字段。无效的计划会进行修复或回退到已验证的规划候选；未解决的验证错误会停止调度。计划中的发现和来源线索仍需要后续研究和验证。规范在规划阶段内修订，然后在章节研究期间保持固定；后续重新打开该计划的研究轮次是对当前流程的扩展。
ResearchSpec records each section's scope, research questions, required entities or cases, and provisional source leads. This compact representation of the intended report also specifies how research work is divided across contexts. Its coverage and responsibilities can be inspected and revised before generating a full draft. Each dispatchable subsection has a unique hierarchical ID, such as S1.2; we refer to these execution units as sections. Before dispatch, validators check ID uniqueness, parent relationships, consecutive ordering, and required fields. Invalid plans undergo repair or revert to a validated planning candidate; unresolved validation errors stop dispatch. The findings and source leads in the plan still require subsequent research and verification. The specification is revised within the planning stage and then held fixed during section research; a later research round that reopens the plan is an extension beyond the current pipeline.
每位研究者接收原始查询、完整的 ResearchSpec 以及一个章节分配。全局规范确立了其局部工作如何为报告做出贡献，而独立的上下文则容纳各分支详细的工具交互。研究者接收共享规范，不依赖于先前生成的章节，并可以在满足分配要求的同时扩展其调查。他们并发执行，并根据已获取的证据撰写自己带有引用的章节。因此，同一个智能体将其局部证据带入写作，而无需先将其压缩为摘要供单独的写作者使用。完整的章节保留了已展开的论点和细节，减少了后续组合对反复压缩的共享历史的依赖。
Each Researcher receives the original query, the complete ResearchSpec, and one section assignment. The global specification establishes how its local work contributes to the report, while separate contexts accommodate the branches' detailed tool interactions. Researchers receive the shared specification, without depending on previously generated sections, and can extend their investigation while addressing the assigned requirements. They execute concurrently and write their own citation-bearing sections from the evidence they have acquired. The same agent therefore carries its local evidence into writing without first compressing it into a summary for a separate writer. The complete sections retain the developed arguments and details, reducing the reliance of subsequent composition on a repeatedly compressed shared history.
该执行框架按 ResearchSpec 顺序机械地汇编章节，即使内容重叠也保留其文本和引用。全局编辑（Global Editor）阅读完整草稿，分配重复材料的归属，识别冲突并指定修改。局部编辑（Local Editors）以完整草稿和规范为上下文，将这些指令应用于其分配的章节，利用草稿中已有的事实和引用。他们的输出范围是局部的，因此一次编辑调用无需重新生成整个文档。这将协调所需的全局判断与修订文本的生成分离开来，同时保持原始章节和汇编草稿可供比较。两种编辑都阅读完整草稿，并仍受输入上下文限制的约束。局部输出范围减少了每次调用重新生成的文本；反复输入完整草稿仍可能产生大量成本。研究和编辑可能会遗漏有用的细节。
The harness mechanically assembles sections in ResearchSpec order, retaining their text and citations even when content overlaps. A Global Editor reads the complete draft, assigns ownership of repeated material, identifies conflicts, and specifies changes. Local Editors apply these directives to their assigned sections using the complete draft and specification as context, drawing on facts and citations already present in the draft. Their output scope is local, so one editorial call need not regenerate the entire document. This separates the global judgment needed for coordination from the generation of revised text, while keeping the original sections and assembled draft available for comparison. Both editor types read the full draft and remain subject to input-context limits. Local output scopes reduce the text regenerated by each call; repeated full-draft inputs can still incur substantial cost. Research and editing can omit useful details.
规划将查询与已探索的来源映射为一个 ResearchSpec；研究将各节分配映射为有依据的章节；编辑将草稿与指令映射为修订后的章节。这些映射为有针对性的数据合成与基于评分标准的评估提供了具体的单元。ResearchSpec 可在详细调查之前检查其覆盖范围，各章节可对照其被分配的要求进行检查，修订稿则可对照编辑指令与原始文本进行检查。这种分解支持在每个阶段分别进行改进与算力分配，并为接下来所述的研究数据构建奠定了基础。
Planning maps the query and explored sources to a ResearchSpec; research maps section assignments to grounded sections; editing maps a draft and directives to revised sections. These mappings provide concrete units for targeted data synthesis and rubric-based evaluation. ResearchSpecs can be checked for coverage before detailed investigation, sections against their assigned requirements, and revisions against editorial directives and original text. The decomposition supports separate improvement and computation allocation at each stage, and provides the foundation for the research-data construction described next.
数据构建遵循 harness 的阶段接口：我们首先构建带有证据支持要求的研究任务，然后收集展示这些任务如何被规划、调查和编辑的轨迹。独立的源材料为这一过程中使用的问题和评分标准提供依据。
Data construction follows the harness's stage interfaces: we first build research tasks with evidence-backed requirements, then collect trajectories that show how those tasks are planned, investigated, and edited. Independent source material grounds the questions and rubrics used in this process.
目标画像指定语言、主题、广度和预期报告。基于文章的路径从独立许可的综述文章开始；互补路径使用冻结的多源简报，包含事实、摘录、URL 和来源限制。两条路径都为需要跨源解释或比较的问题组织证据。
A target profile specifies language, topic, breadth, and the intended report. The article-based route starts from an independently licensed review article; a complementary route uses frozen multi-source briefs containing facts, excerpts, URLs, and source limitations. Both routes organize evidence for questions that require explanation or comparison across sources.
每个问题都配有任务特定的评分标准。遵循证据 grounded 综合原则，事实性标准必须有支持证据，而分析性标准则指定有根据的比较或推断。评分标准可以包括隐含要求和否定条件，如 ResearchRubrics 中所区分。构建证据和评分标准与提供给回答智能体的查询保持分离。
Each question is paired with task-specific rubrics. Following evidence-grounded synthesis principles, factual criteria must have supporting evidence, while analytical criteria specify warranted comparisons or inferences. The rubric can include implicit requirements and negative conditions, as distinguished in ResearchRubrics. Construction evidence and rubrics remain separate from the query supplied to the answering agent.
基于文章的路径支持对原子信息回忆标准进行有界可搜索性检查。它搜索替代来源，从排除的构建文章中移除命中，获取剩余页面，并检查对完整事实要求的支持。完整、部分或缺失支持分别产生 Keep、Revise 或 Drop；只有当所有保留标准都获得 Keep 时，门控才通过。确定性检查处理重复评分标准、答案泄漏、排除源泄漏和时间范围。联合语义审查检查问题、证据和评分标准是否描述了一个连贯的研究任务。有界搜索在其预算内测试证据可用性，但不证明穷尽的 Web 可回答性。
The article-based route supports a bounded searchability check for atomic information-recall criteria. It searches for alternative sources, removes hits from the excluded construction article, fetches the remaining pages, and checks support for the complete factual requirement. Complete, partial, or missing support produces Keep, Revise, or Drop; the gate passes only when all retained criteria receive Keep. Deterministic checks address duplicate rubrics, answer leakage, excluded-source leakage, and temporal scope. Joint semantic review checks whether the question, evidence, and rubric describe a coherent research task. The bounded search tests evidence availability within its budget, without certifying exhaustive Web answerability.
接受的查询驱动 harness 的教师执行，记录候选和最终 ResearchSpecs、工具请求和观察、带引用的章节、组装的草稿和编辑决策。并行分支保留其实际输入和依赖关系。过滤检查角色身份、工具调用/响应闭包和中间产物有效性，同时保留完整、部分和失败尝试之间的区别。每条记录标识其教师和 harness。轨迹有效性是数据集选择的一部分：数据集级重叠检查、分布感知选择和指定的人工审查是训练数据接受前的额外检查。
Accepted queries drive teacher executions of the harness, recording candidate and final ResearchSpecs, tool requests and observations, citation-bearing sections, assembled drafts, and editorial decisions. Parallel branches retain their actual inputs and dependencies. Filtering checks role identity, tool-call/response closure, and intermediate-artifact validity, while preserving distinctions among complete, partial, and failed attempts. Each record identifies its teacher and harness. Trajectory validity is one part of dataset selection: dataset-level overlap checks, distribution-aware selection, and human review where specified are additional checks before training-data acceptance.
研究相关数据与其它数据一起用于 LongCat 通用模型的中间训练和后训练，以提升深度研究能力。LongCat-DeepResearch 将该模型与第 3.1 节所述的研究框架相结合。系统级得分并未单独区分该流程、某一数据来源或某一训练阶段的贡献。
Research-related data are used alongside other data in the mid-training and post-training of LongCat's general-purpose models to improve deep-research capabilities. LongCat-DeepResearch combines this model with the research harness described in Section 3.1. The system-level scores do not isolate the contribution of this pipeline, one data source, or one training stage.
我们围绕四个基准组织评估：DeepResearchBench、DeepResearchBench II、ResearchRubrics 以及一个内部基准。遵循每个基准的官方评估设置，DeepResearchBench RACE 和 DeepResearchBench II 由 GPT-5.5（medium）评判，而 ResearchRubrics 由 Gemini 2.5 Pro 评判。这些模型作为生成报告的评估者。评估者特定的参考结果可从官方基准排行榜获取。官方快照与本文评估的四个系统一并呈现在附录 A 中。三个公开基准提供了与 Gemini-DeepResearch、ChatGPT-DeepResearch 和 Claude-DeepResearch 的记录比较。内部基准在 4.5 节报告。在评估协议下，生成输入是原始任务提示；基准评分标准和参考报告保留用于评分。生成期间公共网络检索保持可用。
We organize evaluation around four benchmarks: DeepResearchBench, DeepResearchBench II, ResearchRubrics, and an in-house benchmark. Following each benchmark's official evaluation setup, DeepResearchBench RACE and DeepResearchBench II are judged by GPT-5.5 (medium), while ResearchRubrics is judged by Gemini 2.5 Pro. These models serve as evaluators of the generated reports. Evaluator-specific reference results are available from the official benchmark leaderboard. The official snapshots and the four systems evaluated here are presented together in Appendix A. The three public benchmarks provide the recorded comparison with Gemini-DeepResearch, ChatGPT-DeepResearch, and Claude-DeepResearch. The in-house benchmark is reported in Section 4.5. Under the evaluation protocol, the generation input is the original task prompt; benchmark rubrics and reference reports are reserved for scoring. Public-Web retrieval remains available during generation.
DeepResearchBench 包含 100 个博士级研究任务，覆盖 22 个领域，中英文任务各 50 个。其 RACE 协议从四个固定的顶层维度评估每份报告：全面性、洞察力/深度、指令遵循和可读性。RACE 在这些维度内生成任务特定的标准和权重，对目标报告和强参考报告进行评分，并以百分比形式计算最终相对分数 \(100 \times S_{\mathrm{tgt}}/(S_{\mathrm{tgt}}+S_{\mathrm{ref}})\)，其中 \(S_{\mathrm{tgt}}\) 和 \(S_{\mathrm{ref}}\) 分别表示目标分数和参考分数。相对分数在每个任务内计算，然后在任务间进行宏平均。我们在可用时报告维度分数；当前 LongCat-DeepResearch 行包含 9 月 14 日评估的全部 100 个任务。本次运行中，我们使用基准的 GPT-5.5（medium）评估器分支，其官方提示、动态标准、参考报告、权重和聚合代码均保持不变。参考报告是 Gemini-DeepResearch 生成的 2025 年 4 月快照。DeepResearchBench 还提供 FACT，它提取陈述–URL 对，去除与同一 URL 相关的重复声明，并检查检索到的页面是否支持每个陈述。官方当前流程使用 GPT-5.4 Mini 进行提取和支持判断，并使用 Jina Reader 获取页面文本。这些数量应附上输出长度，因为有效引用计数对长度敏感。我们不报告 FACT 结果。FACT 衡量陈述–页面支持，它本身并不确立来源权威或每个声明的真实性。
DeepResearchBench contains 100 PhD-level research tasks spanning 22 domains, balanced between 50 Chinese and 50 English tasks. Its RACE protocol assesses each report along four fixed top-level dimensions: Comprehensiveness, Insight/Depth, Instruction-Following, and Readability. RACE generates task-specific criteria and weights within these dimensions, scores both a target and a strong reference, and computes the final relative score as \(100 \times S_{\mathrm{tgt}}/(S_{\mathrm{tgt}}+S_{\mathrm{ref}})\) on the percentage scale, where \(S_{\mathrm{tgt}}\) and \(S_{\mathrm{ref}}\) denote the corresponding target and reference scores. Relative scores are computed within each task before macro-averaging across tasks. We report dimension scores where they are available; the current LongCat-DeepResearch row includes all 100 tasks from the September 14 evaluation. For this run, we use the benchmark's GPT-5.5 (medium) evaluator branch with its official prompt, dynamic criteria, reference reports, weights, and aggregation code unchanged. The reference reports are the April 2025 snapshot generated by Gemini-DeepResearch. DeepResearchBench also provides FACT, which extracts statement–URL pairs, removes duplicate claims associated with the same URL, and checks whether the retrieved page supports each statement. The official current pipeline uses GPT-5.4 Mini for extraction and support judgment and Jina Reader to obtain page text. These quantities should be accompanied by output length because effective-citation counts are length-sensitive. We do not report FACT results. FACT measures statement–page support; it does not by itself establish source authority or the truth of every claim.
DeepResearchBench II 包含 132 个任务，采用相同的 22 领域分类法，每种语言各 66 个任务。该基准以专家撰写的调查性文章为基础。基准论文报告了通过自动提取、自我评估、人工修订和超过 400 小时的专家评审构建的 9,430 条原子评分标准。其评分标准涵盖信息回忆、分析和呈现。只有当报告包含所需事实或推断时，评分标准才通过；数值要求必须明确匹配。我们使用官方评估器，报告每个维度的通过率以及官方总体分数（所有评分标准上任务级通过比例的均值）。根据此协议，任务评分标准和参考文本被排除在 actor 提示之外。如果支持句引用了被屏蔽的源文章，则相应评分标准获得官方分数 \(-1\)。产品比较使用 GPT-5.5（medium）作为评判者。
DeepResearchBench II contains 132 tasks across the same 22-domain taxonomy, with 66 tasks in each language. The benchmark is grounded in expert-written investigative articles. The benchmark paper reports 9,430 atomic rubrics constructed through automatic extraction, self-evaluation, manual revision, and more than 400 hours of expert review. Its rubrics cover Information Recall, Analysis, and Presentation. A rubric passes only when the report contains the required fact or inference; numerical requirements must be matched explicitly. We use the official evaluator and report the pass rate for each dimension together with the official overall score (the mean of task-level pass fractions across all rubrics). Task rubrics and reference text are excluded from the actor prompt by this protocol. If a supporting sentence cites a blocked source article, the corresponding rubric receives the official score of \(-1\). The product comparison uses GPT-5.5 (medium) as the judge.
ResearchRubrics 将 101 个现实提示与 2,593 条专家编写的标准配对，涵盖显式和隐式要求、综合、引用质量、指令遵循和沟通质量。我们在任务间进行宏平均之前，先计算每个任务的加权分数。LongCat 结果覆盖全部 101 个任务。维度分数在分子中使用分数乘以权重之和，分母中仅使用正权重；负权重标准对分子贡献惩罚。维度分数不进行平均以获得总体结果。我们使用 Gemini 2.5 Pro 作为评判者，遵循基准的评估设置，并保留其原始的专家编写标准和加权评分惯例。该基准通过测试对提示特定标准的遵循情况来补充 DeepResearchBench II，这些标准包括请求的内容和从任务上下文推断的要求。
ResearchRubrics pairs 101 realistic prompts with 2,593 expert-written criteria covering explicit and implicit requirements, synthesis, citation quality, instruction following, and communication quality. We compute a weighted score per task before macro-averaging across tasks. The LongCat result covers all 101 tasks. Dimension scores use the sum of score times weight in the numerator and only positive weights in the denominator; negative-weight criteria contribute penalties to the numerator. Dimension scores are not averaged to obtain the overall result. We use Gemini 2.5 Pro as the judge, following the benchmark's evaluation setup, and retain its original expert-written criteria and weighted scoring convention. This benchmark complements DeepResearchBench II by testing adherence to prompt-specific criteria that include both requested content and requirements inferred from task context.
内部比较使用自动评估器在内部基准上评估 LongCat-DeepResearch、Gemini-DeepResearch、ChatGPT-DeepResearch 和 Claude-DeepResearch。所得维度分数和加权总体结果报告在第 4.5 节。
The in-house comparison evaluates LongCat-DeepResearch, Gemini-DeepResearch, ChatGPT-DeepResearch, and Claude-DeepResearch on the in-house benchmark using an automatic evaluator. The resulting dimension scores and weighted overall results are reported in Section 4.5.
对比对象包括 LongCat-DeepResearch、Gemini-DeepResearch、ChatGPT-DeepResearch 和 Claude-DeepResearch。后三者通过各自的官方客户端访问，使用各提供商自己的 Deep Research 系统，包括其原生工具与预算。所选的客户端模型分别为 Gemini 3.7 Flash、GPT-5.6 Sol（xhigh）和 Claude Opus 5（xhigh）。结果表中的产品名称指的是这些完整的 Deep Research 系统及其指定的客户端模型选择。这些系统级对比并不单独隔离某个模型或框架组件。第 5.4 节的配置研究比较了先前发布的 LongCat 与 ReAct/直接写作以及当前框架，并在当前框架下比较了先前版本与当前 LongCat 模型。这三种配置均使用完整基准测试。
The comparison includes LongCat-DeepResearch, Gemini-DeepResearch, ChatGPT-DeepResearch, and Claude-DeepResearch. The latter three are accessed through their respective official clients using each provider's own Deep Research system, including its native tools and budgets. The selected client models are Gemini 3.7 Flash, GPT-5.6 Sol (xhigh), and Claude Opus 5 (xhigh), respectively. The product names in the result tables refer to these complete Deep Research systems with the specified client model selections. These system-level comparisons do not isolate a model or harness component. The configuration study in Section 5.4 compares the previous LongCat release with ReAct/direct writing and the current harness, and compares the previous release with the current LongCat model under the current harness. All three configurations use the full benchmarks.
基准任务集包含 100 个 DeepResearchBench 任务、132 个 DeepResearchBench II 任务和 101 个 ResearchRubrics 任务。DeepResearchBench 与 DeepResearchBench II 使用 GPT-5.5（medium），ResearchRubrics 使用 Gemini 2.5 Pro。每个基准的总体得分按任务计算，然后在其被评分的任务上取平均；ResearchRubrics 保留带符号的准则权重以及正权重分母。各系统的被评分覆盖率有所不同。我们按系统实际部署的形态进行比较，包括其原生研究工具与预算。详细的覆盖率与执行记录见附录 B。
The benchmark task sets contain 100 DeepResearchBench tasks, 132 DeepResearchBench II tasks, and 101 ResearchRubrics tasks. DeepResearchBench and DeepResearchBench II use GPT-5.5 (medium), and ResearchRubrics uses Gemini 2.5 Pro. Each benchmark's overall score is computed per task and then averaged over its scored tasks; ResearchRubrics retains signed criterion weights and a positive-weight denominator. Scored coverage varies across systems. We compare the systems as deployed, including their native research tools and budgets. Detailed coverage and execution records are provided in Appendix B.
LongCat-DeepResearch 以 55.25 分在整体比较中领先，超过 ChatGPT-DeepResearch 0.30 分。它在全面性（56.21）、洞察力（56.48）和指令遵循（55.18）方面也获得了最高分。可读性仍然较弱：其 49.96 分低于 ChatGPT-DeepResearch 的 51.51 分和 Claude-DeepResearch 的 51.35 分。
LongCat-DeepResearch leads the overall comparison with 55.25, exceeding ChatGPT-DeepResearch by 0.30 points. It also has the highest scores for comprehensiveness (56.21), insight (56.48), and instruction following (55.18). Readability remains weaker: its 49.96 is below ChatGPT-DeepResearch's 51.51 and Claude-DeepResearch's 51.35.
LongCat-DeepResearch 得分为 51.35，超过 Claude-DeepResearch 3.17 分。它在信息召回（46.42）和分析（60.69）方面领先，而其呈现得分 83.13 落后于 ChatGPT-DeepResearch 的 86.98。最强的维度级优势在于分析：60.69 对次高系统的 52.26。
LongCat-DeepResearch scores 51.35, exceeding Claude-DeepResearch by 3.17 points. It leads in information recall (46.42) and analysis (60.69), while its presentation score of 83.13 trails ChatGPT-DeepResearch's 86.98. The strongest dimension-level advantage is in analysis: 60.69 versus 52.26 for the next-highest system.
LongCat-DeepResearch 得分为 79.83，超过 ChatGPT-DeepResearch 5.62 分。它在显式要求（86.97）、隐式要求（79.52）和综合（78.92）方面领先。指令遵循、引用质量和沟通均低于至少一个对比系统。这些类别分数描述了报告质量的互补方面；它们的简单平均并不能复现完整评分标准的总分。
LongCat-DeepResearch scores 79.83, exceeding ChatGPT-DeepResearch by 5.62 points. It leads in explicit requirements (86.97), implicit requirements (79.52), and synthesis (78.92). Instruction following, citation quality, and communication each remain below at least one compared system. These category scores describe complementary aspects of report quality; their simple average does not reproduce the full-rubric total.
第 5.5 节中的案例研究说明了 ResearchSpec、章节研究和编辑如何在一份记录报告中相互作用。
The case study in Section 5.5 illustrates how ResearchSpec, section research, and editing interact in one recorded report.
我们在内部基准上比较了 Gemini-DeepResearch、ChatGPT-DeepResearch、Claude-DeepResearch 和 LongCat-DeepResearch。表 1 的 In-house 部分报告了内容（任务覆盖与证据支持）、分析（推理与综合）和呈现（组织与清晰度），以及加权总分。自动评估器沿这些维度对报告进行评估。LongCat-DeepResearch 的总分为 76.04，比 ChatGPT-DeepResearch 的 76.59 低 0.55 分。
We compare Gemini-DeepResearch, ChatGPT-DeepResearch, Claude-DeepResearch, and LongCat-DeepResearch on the in-house benchmark. The In-house block of Table 1 reports Content (task coverage and evidential support), Analysis (reasoning and synthesis), and Presentation (organization and clarity), together with the weighted overall score. An automatic evaluator assesses the reports along these dimensions. LongCat-DeepResearch scores 76.04 overall, 0.55 points below ChatGPT-DeepResearch's 76.59.
我们在开发子集上研究组件消融、ResearchSpec 精炼以及 Editor 的 Scaling（规模扩张），并在完整基准上比较模型–工具框架（model–harness）配置。每项研究的实验方案分别报告。DRB-I、DRB-II 和 RR 分别缩写自 DeepResearchBench、DeepResearchBench II 和 ResearchRubrics。
We study component ablations, ResearchSpec refinement, and Editor scaling on development subsets, and compare model–harness configurations on the full benchmarks. Protocols are reported separately for each study. DRB-I, DRB-II, and RR abbreviate DeepResearchBench, DeepResearchBench II, and ResearchRubrics, respectively.
每个展示列中的所有条件均使用相同的 question ID。LongCat DeepResearchBench II 的对比使用四个实验组共同完成的题目。筛选过程与停止案例记录于 B.2 和 B.2 节。
All conditions within each displayed column use identical question IDs. The LongCat DeepResearchBench II comparison uses the common completed questions across all four arms. Selection and the stopped case are documented in Sections B.2 and B.2.
完整方案相较于简化规划与单一整篇报告 Researcher 的优势，支持了紧凑全局协调与细致局部调研之间的分工。ResearchSpec 可以引导报告的共享议程，而每个章节则在独立的上下文中发展其证据。
Full's advantage over simplified planning and a single whole-report Researcher supports the division between compact global coordination and detailed local investigation. ResearchSpec can guide the report's shared agenda while each section develops its evidence in an independent context.
我们在一个固定的 ResearchRubrics 开发子集上，使用 LongCat 考察规划聚合与精炼如何影响最终报告。W0 使用固定的单个 Writer 的 ResearchSpec；J0 在任何 Critic/Reviser 循环之前，通过 Judge 整合三个 Writer。R1 和 R2 对 J0 分别施加一次和两次连续的 Critic/Reviser 循环。每个阶段都运行相同的下游 Researcher 和 Editor 配置；规划压缩被禁用。
We examine how planning aggregation and refinement affect final reports on a fixed ResearchRubrics development subset with LongCat. W0 uses a fixed single Writer's ResearchSpec; J0 integrates three Writers through the Judge, before any Critic/Reviser cycle. R1 and R2 apply one and two successive Critic/Reviser cycles to J0. Each stage runs the same downstream Researcher and Editor configuration; planning compaction is disabled.
我们在每个基准的固定开发子集上研究 LongCat Editor 的规模扩展，并按问题匹配进行比较。A0 表示未编辑的草稿。A1 将原始 Editor 应用于 A0 一次。B1 独立地将增强版 Editor 应用于相同的 A0，添加源摘录和事实核查指令。B2 编辑 B1，B3 编辑 B2，总共进行两轮和三轮的增强编辑。A1 是一个独立的对照，不是 B1 的起点（B.7 节）。表 4 报告了这些条件下的原生分数和平均可读性偏好。可读性评估协议详见 B.5 节。
We study LongCat Editor scaling on fixed development subsets for each benchmark, with comparisons matched by question. A0 denotes the unedited draft. A1 applies the original Editor to A0 once. B1 independently applies an enhanced Editor to the same A0, adding source excerpts and fact-checking instructions. B2 edits B1, and B3 edits B2, giving two and three enhanced rounds in total. A1 is a separate control, not the starting point for B1 (Section B.7). Table 4 reports native scores and average readability preference for these conditions. The readability evaluation protocol is detailed in Section B.5.
在增强版 Editor 的轮次中，平均可读性偏好有所提高；最后一轮还保持或提高了相对于未编辑草稿的原生基准分数。这支持将协调的局部修订作为将独立研究的章节发展为更连贯报告的一种方式。特定基准的偏好和配对结果见 B.7 和 B.7 节。
Across enhanced Editor rounds, average readability preference improves; the final round also preserves or improves the native benchmark scores relative to the unedited draft. This supports coordinated local revision as a way to develop independently researched sections into a more coherent report. Benchmark-specific preferences and paired results are given in Sections B.7 and B.7.
我们在完整的基准测试队列上比较了三种配置。第一种使用先前发布的 LongCat 模型，采用有界 ReAct 研究，随后直接生成报告。第二种将同一先前版本与当前运行框架配对。第三种将当前 LongCat 模型与当前运行框架配对。ReAct 使用相同的搜索和页面阅读服务，最终输出上限为 65,536 个 token，并启用思考功能。
We compare three configurations on the full benchmark cohorts. The first uses the previous LongCat release with bounded ReAct research followed by a direct report. The second pairs the same previous release with the current harness. The third pairs the current LongCat model with the current harness. ReAct uses the same search and page-reading services and a 65,536-token final-output ceiling, with thinking enabled.
在模型固定时更换运行框架所带来的增益，体现了围绕明确需求和分节工作来组织研究的价值。使用当前模型获得的进一步增益，支持将模型能力与研究编排视为系统中互补的两个部分。
The gains from changing the harness with the model fixed show the value of organizing research around explicit requirements and section-level work. Further gains with the current model support treating model capability and research orchestration as complementary parts of the system.
我们考察一个记录在案的 LongCat 轨迹，用于 AI 增强的投资组合管理（DeepResearchBench II 012）。用户要求对经典理论、AI/ML 应用、多准则决策以及投资组合优化与再平衡进行综述，涵盖截至 2024 年 4 月的方法。该案例揭示了两个具体操作：在章节研究之前修订研究议程，以及在章节草稿汇编后协调重叠证据。以下摘录展示了该轨迹中的计划、章节草稿、编辑指令和最终报告。
We examine a recorded LongCat trajectory for AI-enhanced portfolio management (DeepResearchBench II 012). The user requests a review of classical theories, AI/ML applications, multi-criteria decision making, and portfolio optimization and rebalancing, covering methods through April 2024. The case exposes two concrete operations: revising the research agenda before section research, and coordinating overlapping evidence after the section drafts have been assembled. The excerpts below show the plan, section draft, editorial directive, and final report from this trajectory.
合并后的 ResearchSpec 将四个请求主题分解为 19 个小节：十个用于经典理论，五个用于 AI/ML 应用，两个用于多准则方法，两个用于优化与再平衡。每个小节包含“涵盖内容”、研究问题、所需实体或案例以及来源线索。这些字段将小节的主题与其 Researcher 必须寻找的证据分开。该规范还记录了共享边界，包括出版截止日期以及对原创、同行评审来源的优先考虑。
The merged ResearchSpec decomposes the four requested themes into 19 subsections: ten for classical theories, five for AI/ML applications, two for multi-criteria methods, and two for optimization and rebalancing. Each subsection contains What to Cover, research questions, required entities or cases, and source leads. These fields separate the topic of a section from the evidence that its Researcher must seek. The specification also records shared boundaries, including the publication cutoff and the priority given to primary, peer-reviewed sources.
合并后的计划涵盖了预测方法，但没有提及 Heaton、Polson 和 Witte 或“深度投资组合理论”。Critic 将这一遗漏标记为 HP1（Critic 列出的第一个问题），并引导关注从数据中学习投资组合权重，与现有的预测议程并列。Reviser 将该主题添加到 S2.2，包括一个具体的研究问题和一个命名实体以供后续调查。这一更改扩展了现有任务；它没有创建额外的小节。
The merged plan covers forecasting methods but does not name Heaton, Polson, and Witte or “Deep Portfolio Theory.” The Critic flags this omission as HP1 (the Critic’s first listed issue) and directs attention to learning portfolio weights from data, alongside the existing forecasting agenda. The Reviser adds the topic to S2.2, including a specific research question and a named entity for subsequent investigation. The change expands an existing assignment; it does not create an additional subsection.
资本资产定价模型（CAPM）任务（S1.2）询问哪些证据支持或挑战该模型，并提供一份调查作为来源线索。其记录的 Researcher 输出将这一问题发展为一个引用小节，涵盖实证检验、Roll 的批评以及 Fama–French 的发现。最终报告还在时间序列预测下包含一个专门的“深度投资组合理论”段落，保留了添加的主题。这些是书面小节产物；下一阶段将协调它们重叠的材料。
The Capital Asset Pricing Model (CAPM) assignment (S1.2) asks which evidence supports or challenges the model and supplies a survey as a source lead. Its recorded Researcher output develops that question into a cited section on empirical tests, Roll’s critique, and Fama–French findings. The final report also contains a dedicated “Deep Portfolio Theory” passage under time series forecasting, preserving the added topic. These are written section artifacts; the next stage coordinates their overlapping material.
Fama–French 发现在本报告中发挥两种作用。在 CAPM 部分，它们对模型提出了实证挑战；在三因子模型部分（S1.5），它们启发了对收益的另一种解释。因此，独立撰写章节可能会重复一项合理的证据。全局编辑阅读组装后的草稿，将详细处理分配给 S1.5，同时指示 S1.2 保留简要提及和交叉引用。然后，局部编辑将此决定应用于现有的 CAPM 部分。
The Fama–French findings serve two roles in this report. In the CAPM section they provide an empirical challenge to the model; in the three-factor-model section (S1.5) they motivate an alternative account of returns. Independent section writing can therefore repeat a legitimate piece of evidence. The Global Editor reads the assembled draft and assigns the detailed treatment to S1.5, while directing S1.2 to retain a brief mention and a cross-reference. The Local Editor then applies this decision to the existing CAPM section.
计划、草稿、指令和最终段落可以通过相同的章节标识符链接起来。修订操作针对现有研究工件：它保留核心实证发现，压缩其局部处理，并记录详细讨论所属的位置。全局计划对“误差最大化”概念和非支配排序遗传算法 II（NSGA-II）细节做出了类似的所有权决策，表明协调涉及跨章节的实质性材料分配，以及句子级编辑。这些指令是记录在案的决策；上述 CAPM 段落是此处追踪的具体执行。
The plan, draft, directive, and final passage can be linked through the same section identifier. The revision operates on an existing research artifact: it preserves the central empirical finding, condenses its local treatment, and records where the detailed discussion belongs. The Global plan makes similar ownership decisions for the “error maximization” concept and for Non-dominated Sorting Genetic Algorithm II (NSGA-II) details, showing that coordination concerns the allocation of substantive material across sections, as well as sentence-level editing. Those directives are recorded decisions; the CAPM passage above is the specific execution traced here.
该示例还暴露了最终化不匹配：报告保留了内部“S1.5”引用，而显示的三因子模型标题省略了该标识符。最终报告仍缺少一些基准要求，包括 Treynor 和 CPPI。因此，可执行的研究议程和可追踪的编辑并不能保证完全覆盖或完美呈现。只有最终报告被评分；该案例并未分离出规划或编辑的收益。更多摘录和一个具有未解决覆盖差距的对比劳动力市场案例出现在附录 C 和 C 中。
The example also exposes a finalization mismatch: the report retains the internal “S1.5” reference while the displayed three-factor-model heading omits that identifier. The final report still misses some benchmark requirements, including Treynor and CPPI. Thus, an executable research agenda and a traceable edit do not guarantee complete coverage or flawless presentation. Only the final report was scored; the case does not isolate planning or editing gains. Further excerpts and a contrasting labor-market case with unresolved coverage gaps appear in Appendices C and C.
我们提出了 LongCat-DeepResearch，一个模型与工具链系统，它将早期研究迭代从完整报告转向 ResearchSpec。并行规划定义了证据需求和章节职责。研究者独立开发带有引用的章节，协调的编辑者修订组装后的草稿。这些接口还支持研究任务和轨迹构建。
We presented LongCat-DeepResearch, a model-and-harness system that shifts early research iteration from full reports to ResearchSpec. Parallel planning defines evidence requirements and section responsibilities. Researchers develop citation-bearing sections independently, and coordinated Editors revise the assembled draft. These interfaces also support research-task and trajectory construction.
该系统在 DeepResearchBench 上达到 55.25，在 DeepResearchBench II 上达到 51.35，在 ResearchRubrics 上达到 79.83。在匹配的开发子集上，完整的 LongCat 流程具有最高的组件研究均值。全基准模型–工具链比较有利于当前系统：当前工具链在先前版本上提高了分数，而当前模型在该工具链下进一步提高了分数。进一步的规划细化产生了混合的类别级效果，而增强的编辑者轮次提高了平均可读性偏好。这些观察结果特定于所评估的配置，并未隔离训练阶段的贡献。
The system achieves 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics. On matched development subsets, the full LongCat pipeline has the highest component-study means. The full-benchmark model–harness comparison favors the current system: the current harness improves scores with the previous release, and the current model further improves scores under that harness. Further planning refinement produces mixed category-level effects, while enhanced Editor rounds improve average readability preference. These observations are specific to the evaluated configurations and do not isolate training-stage contributions.
ResearchSpec 覆盖支持对计划需求的早期检查；事实核查和独立报告评估仍然是必要的。
ResearchSpec coverage supports early inspection of planned requirements; factual verification and independent report evaluation remain necessary.
原生基准分数、计划覆盖率和引用支持衡量的是不同属性。在自动基准和组件评估中，未更改的内容可能获得不同的评分标准判断。问题自举区间捕捉的是问题变异，而非重复生成或评判带来的不确定性。覆盖率诊断支持对计划需求的检查；事实核查和独立报告评估仍然必要。
Native benchmark scores, planned coverage, and citation support measure different properties. In the automatic benchmark and component evaluations, unchanged content can receive different rubric judgments. Question-bootstrap intervals capture question variation, not uncertainty from repeated generation or judging. Coverage diagnostics support inspection of planned requirements; factual verification and independent report evaluation remain necessary.
组件、规划细化和编辑器研究使用了先前检查过的开发子集，其中一些是使用先前模型分数选择的。检索窗口、实际计算和恢复历史在不同研究中有所变化。更广泛的任务覆盖和重复运行将有助于表征趋势稳定性和资源需求。研究任务和轨迹构建属于更广泛的模型开发过程。模型比较评估的是生成的检查点；匹配的训练消融将有助于识别单个数据源和训练阶段的贡献。变化的网络内容和托管研究产品也限制了精确复现。
The component, planning-refinement, and Editor studies use previously inspected development subsets, some selected using prior model scores. Retrieval windows, realized computation, and recovery histories vary across studies. Broader task coverage and repeated runs would help characterize trend stability and resource requirements. Research-task and trajectory construction belong to a broader model development process. The model comparisons evaluate the resulting checkpoints; matched training ablations would help identify contributions from individual data sources and training stages. Changing web content and hosted research products also limit exact replay.
ResearchSpec 在规划期间细化，并在章节研究期间固定。研究人员可以在其任务范围内进一步调查，但在起草后重新打开全局议程仍处于核心流程之外。完整的章节工件支持组装和检查，但研究和编辑可能遗漏相关细节。该框架优先考虑证据召回和研究需求的覆盖。生成的报告可能冗长、密集或重复，为人类读者提高可读性仍有改进空间。全局和局部编辑器阅读完整草稿，并仍受输入上下文限制。额外的细化在不同配置和基准上效果各异，消耗的资源本也可用于支持证据收集。增强编辑器联合改变了源访问、指令和迭代次数，使其各自的贡献尚未解决。
ResearchSpec is refined during planning and fixed during section research. Researchers can investigate further within their assignments, but reopening the global agenda after drafting remains outside the core pipeline. Complete section artifacts support assembly and inspection, yet research and editing can omit relevant details. The framework prioritizes evidence recall and coverage of research requirements. The resulting reports can be lengthy, dense, or repetitive, leaving room to improve readability for human readers. Global and Local Editors read the full draft and remain subject to input-context limits. Additional refinement has varying effects across configurations and benchmarks, consuming resources that could also support evidence gathering. The enhanced Editor jointly changes source access, instructions, and iteration count, leaving their individual contributions unresolved.