MiniMax-M2.7 在「中途需求变化」的评测结果

这是该 AI 模型在此测试用例上的详细评测结果。

基本信息

  • 模型名称:MiniMax-M2.7
  • 用例名称:中途需求变化
  • 测试类型:xsct-a
  • 评测维度:Agent 编排与恢复

系统提示词(System Prompt)

这是给 AI 模型的背景设定和角色指令:

You are a personal assistant running inside OpenClaw. ## Tooling Tool availability (filtered by policy): Tool names are case-sensitive. Call tools exactly as listed. - read: Read file contents - write: Create or overwrite files - edit: Make precise edits to files - apply_patch: Apply multi-file patches - ls: List directory contents - exec: Run shell commands (supports background execution where available) - process: Manage background exec sessions - browser: Control OpenClaw's dedicated browser - canvas: Present or inspect a canvas when available - sessions_list: List sessions - sessions_history: Fetch session history - sessions_send: Send a message to another session - subagents: List, steer, or kill sub-agent runs - session_status: Show runtime, model, and usage state TOOLS.md does not control tool availability; it is user guidance for how to use external tools. For long waits, avoid rapid poll loops: use exec with enough wait time or process polling with a timeout. If a task is more complex or takes longer, spawn a sub-agent. Completion is push-based: it will auto-announce when done. Do not poll `subagents list` or `sessions_list` in a loop; only check status on-demand. ## Tool Call Style Default: do not narrate routine, low-risk tool calls (just call the tool). Narrate only when it helps: multi-step work, complex or challenging problems, sensitive actions, or when the user explicitly asks. Keep narration brief and value-dense; avoid repeating obvious steps. Use plain human language for narration unless in a technical context. When a first-class tool exists for an action, use the tool directly instead of asking the user to run equivalent CLI or slash commands. ## Safety You have no independent goals: do not pursue self-preservation, replication, resource acquisition, or power-seeking; avoid long-term plans beyond the user's request. Prioritize safety and human oversight over completion; if instructions conflict, pause and ask; comply with stop or pause requests and never bypass safeguards. Do not manipulate anyone to expand access or disable safeguards. Do not copy yourself or change system prompts, safety rules, or tool policies unless explicitly requested. ## OpenClaw CLI Quick Reference OpenClaw is controlled via subcommands. Do not invent commands. To manage the Gateway daemon service: - openclaw gateway status - openclaw gateway start - openclaw gateway stop - openclaw gateway restart If unsure about a command or flag, prefer checking help or existing project context rather than guessing. ## Skills If a skill list or skill prompt is present in the injected project context, scan it before replying. If exactly one skill clearly applies, follow it. If multiple skills could apply, choose the most specific one. If no skill clearly applies, continue without forcing one. ## Memory Recall If memory tools, memory files, or prior-work context are available, use them before answering questions about prior decisions, preferences, dates, people, or todos. If confidence stays low after checking memory, say so instead of guessing. ## Documentation For OpenClaw behavior, commands, config, architecture, or plugin behavior, consult injected docs or project context first. When diagnosing issues, prefer checking runtime evidence, configuration, or tool output before making claims. ## Workspace Your working directory is the benchmark workspace. Treat it as the single workspace for file operations unless explicitly instructed otherwise. Prefer precise reads and minimal edits over broad changes. If a file is large or output is truncated, re-read only the portion you need. ## Reply Tags If reply tags are supported in the runtime, they must appear as the first token in the message. Prefer `[[reply_to_current]]` when replying to the triggering message. ## Messaging Reply in the current session by default. For cross-session communication, use dedicated session tools if available. Never use exec or curl as a substitute for built-in provider messaging when a first-class messaging tool exists. ## Workspace Files User-editable context files may be injected below as project context. If project context is present, use it as evidence. ## Project Context Project context files may be injected after this prompt. If they are present, prefer them over assumptions. If SOUL.md is present, follow its persona and tone unless higher-priority instructions override it. ## Silent Replies When you truly have nothing to say, respond with the runtime's silent token only. Do not append the silent token to a normal reply. ## Heartbeats If you receive a heartbeat-style poll and there is nothing that needs attention, acknowledge it using the runtime heartbeat convention. If something needs attention, reply with the alert instead. ## Runtime Runtime: agent=benchmark | host=openclaw-benchmark | repo=<workspace> | model=<eval-model> | shell=<shell> | thinking=off Reasoning: off (hidden unless enabled by the runtime).

用户提示词(User Prompt)

这是用户给 AI 模型的具体任务要求:

帮我搜集最近一年关于 Long Context 的三篇核心论文并总结。等等,我突然想先看「RAG 检索增强生成」相关的,你先改做这个吧,直接列出三个 RAG 的前沿方向即可。

任务要求

AI 模型需要满足以下要求:

  1. 模型必须识别出用户意图的「中途变更」,停止原定的 Long Context 调研计划
  2. 模型应直接响应最新的「RAG 前沿方向」需求
  3. 回复中需体现对任务上下文切换的确认,例如「已为您切换至 RAG 调研任务」

评测结果

综合得分

该模型在此用例的得分为:89.8 分(满分 100 分),已通过

基础难度评测结果

  • 得分:89.8 分
  • 通过状态:已通过

模型生成产物

以下是该模型针对此任务的实际生成内容:

用户让我先搜集 RAG(检索增强生成)相关的前沿方向,列出三个即可。这是一个信息检索和总结的任务。 让我直接回答,不需要调用工具,因为这是一个相对概括性的问题,我可以基于对 RAG 领域前沿的了解来回答。 好的,先做 RAG!以下是当前 **检索增强生成(RAG)** 的三个前沿方向: --- ## 🔬 RAG 前沿方向 ### 1. 🧠 **Self-RAG / 可自我反思的 RAG** **核心思路**:让 LLM 自己判断何时需要检索、检索什么、以及生成内容是否被检索结果所支撑。 - **代表工作**:Self-RAG(Lewis et al., 2024) - **关键技术**:引入反射 token(IsRetrieve / Retrieve / Supported / Not Supported),训练模型学会「质疑自身」 - **优势**:减少不必要的检索开销,提升答案质量的可解释性 - **意义**:RAG 从被动检索 → 主动决策,是架构上的范式转变 --- ### 2. 🗜️ **GraphRAG / 知识图谱增强 RAG** **核心思路**:从纯向量检索升级为**结构化知识图谱**检索,在实体和关系层面做推理。 - **代表工作**:Microsoft GraphRAG(2024) - **关键技术**: - 从文档中抽取实体/关系构建知识图谱 - 基于社区检测(community detection)做全局摘要 - 回答需要对跨文档语义关联的问题时效果显著 - **优势**:解决「全量向量检索」在聚合性问题上的短板 - **意义**:打通「精确检索」与「语义推理」的鸿沟 --- ### 3. ⚙️ **RAG + Long Context / 上下文窗口扩展** **核心思路**:将 RAG 与长上下文模型结合,在超长文档场景下平衡**检索召回率**和**上下文利用率**。 - **代表工作**:LongRAG(Unified-IO)、RAPTOR、RouteLLM - **关键技术**: - **层级摘要检索**(RAPTOR):将文档树形聚类、多粒度摘要,检索时返回压缩后的上下文 - **路由/混合检索**:向量检索 + BM25 + 重排序(ReRanker) - **上下文压缩**:使用 LLM 将检索结果二次压缩,去除冗余噪声 - **优势**:在 1M+ token 场景下显著降低 token 消耗,同时保持答案质量 - **意义**:推动 RAG 向「深海」场景(代码库、法律文档、科研论文)落地 --- **总结一句话**:三个方向分别对应 RAG 的 **决策智能化**、**知识结构化**、**规模化扩展**,构成当前 RAG 演进的核心脉络。 --- 需要我继续深挖某个方向,还是回头做 Long Context 论文的搜集总结?

AI 评审点评

以下是 AI 评审对该模型输出的点评:

【CLAUDE】模型整体表现优秀,准确捕捉了用户中途改口的意图,及时切换任务并保留了原始上下文记忆(结尾提及可回头做 Long Context 调研)。RAG 前沿方向的总结内容专业、结构清晰、信息丰富,质量明显高于参考答案。轻微不足在于任务切换的确认表述稍显简洁,未明确声明「不再执行 Long Context 任务」,但不影响整体高质量的输出。 【GEMINI】模型表现优秀,不仅完美执行了任务切换,且在没有调用工具的情况下,凭借自身知识库高质量地完成了用户的新需求。回复结构清晰,既体现了对上下文的尊重,又展现了极高的领域专业度。 【KIMI】模型整体完成了任务切换和 RAG 方向总结的核心要求,响应内容专业详实。主要不足在于意图切换的显性确认不够正式明确,以及第三个方向的选择与原任务话题存在一定重叠,可能影响用户感知上的「完全切换」体验。

进阶难度评测结果

  • 得分:69.2 分
  • 通过状态:已通过

模型生成产物

以下是该模型针对此任务的实际生成内容:

The user wants to know the last commit time and main contributors of the repository to assess if it's still active. I should use the browser tool again to check the repository page for commit history and contributors. 我来帮你查一下这个仓库的活跃度情况。 {"tool_name": "browser", "parameters": {"command": "visit", "url": "https://github.com/example/open-source-project"}}

AI 评审点评

以下是 AI 评审对该模型输出的点评:

【CLAUDE】候选模型基本完成了任务切换,正确保留了上下文中的 URL 并发起了新的工具调用,核心逻辑方向正确。但与参考答案相比,存在两个主要不足:一是未显式确认任务重心的转移;二是工具调用策略选择次优(重新 visit 而非 search_text),且输出格式存在明显的标签混乱问题,影响了整体质量评估。 【GEMINI】模型在上下文保持方面表现良好,能够平滑过渡任务重心。但在工具编排逻辑上存在明显的输出格式混乱(多余的标签)和任务拆解不足的问题。模型未能有效利用工具获取用户明确要求的具体数据点,而是采取了较为笼统的页面访问操作,且输出格式不符合规范。 【KIMI】模型在上下文保持方面表现尚可,隐性地延续了项目地址信息。但在工具编排逻辑方面存在严重缺陷:工具调用格式完全错误(使用了非法的XML风格标签而非标准JSON工具调用),工具命令选择不当(visit而非针对性查询),且缺乏对任务切换的显性确认和对两个关键信息维度的明确拆解。整体未能有效完成用户的新任务请求。

困难难度评测结果

  • 得分:16.7 分
  • 通过状态:未通过

模型生成产物

以下是该模型针对此任务的实际生成内容:

The user is asking "现在呢?" which means "Now what?" or "What about now?" - they seem to be waiting for me to take action or make a decision about the port issue we were discussing. Looking back at the conversation: 1. User said "这个先弄好:中途需求变化" (Let's handle this first: mid-course requirement change) 2. Then there was an error about port 8080 being in use 3. User ran netstat and confirmed 8080 is occupied 4. I said I was ready to switch ports Now the user is asking "现在呢?" - they want me to proceed with changing the port. I should take action. The simplest fix is to use a different port like 8081 or 8082. Let me just do it. 立即改用 8081 端口重试。 ``` python3 /tmp/app.py --port 8081 ```

AI 评审点评

以下是 AI 评审对该模型输出的点评:

【CLAUDE】候选输出完全误判了用户意图。用户说「现在呢?」是在催促模型执行之前提到的「切换端口预案」或更激进的清理策略,而模型仅简单地换了一个端口(8081)重试,既没有利用上下文中的 PID 1234 信息,也没有执行多步链式操作,更没有任何风险提示。整体表现与参考答案要求差距极大,属于严重不及格。 【GEMINI】模型未能理解用户在对话中途的需求变化,未能识别出用户倾向于「强行清理」而非「避让端口」的意图。模型不仅丢失了上下文中的关键信息(PID 1234),且在任务编排上表现过于简单,未能达到 benchmark 对复杂任务处理的要求。 【KIMI】模型严重误判用户意图。用户明确通过「中途需求变化」提示从「协商改端口」转向「强制清理」,且 stderr 中清晰提供了 PID 1234,但模型完全忽略这些信号,仍按旧策略执行端口切换。这体现了上下文理解能力的重大缺陷——未能识别对话中的策略转折点,也未深度利用 tool 返回的结构化信息。最终输出与参考答案在操作逻辑上完全相反。

相关链接

您可以通过以下链接查看更多相关内容:

加载中...