{ "schemaVersion": 3, "methodologyVersion": "1.2", "snapshotType": "update", "cutoff": "2026-10-16T09:00:00+08:00", "previousCutoff": "2026-10-01T09:00:00+08:00", "researchedAt": "2026-10-16T18:06:00+08:00", "publishedAt": "2026-10-16T18:06:00+08:00", "backfilled": false, "originalScheduledDate": null, "score": null, "band": null, "scoreStatus": "insufficient-evidence", "coverage": 0.0075, "coverageLabel": "评分证据仍只覆盖 0.75%", "calculation": { "formula": "100 / 3 * sum(weight * score)", "weightedSubtotal": -0.75, "note": "仅供算术校验的已知项小计，不是综合分。覆盖不足40%不发布分数或方向。" }, "factualDelta": 0, "methodEffect": null, "coverageDelta": 0, "changeExplanation": "10 月上半月出现更低价的美国小模型、更具体的智能体越界行为回执，以及中国全产业链评测基准。它们分别改善了成本、安全透明度和评测工具，但都没有给出中美同口径原值；总分和区间继续留空。", "level3": [ { "id": "U1.1", "name": "前沿任务综合成功率", "weight": 0.07, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 1, "missingReason": "五类榜单都已复核，但未取得四类任务、两国各前三基础模型家族且统一成功率的完整原表。" }, { "id": "U1.2", "name": "长程任务可靠性", "weight": 0.042, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "U1.3", "name": "可信与安全部署能力", "weight": 0.028, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "Anthropic 披露了非预期网站操作与缓解措施，但没有美中同一对抗、越权和拒答协议下的对称结果。" }, { "id": "U2.1", "name": "每个成功任务的有效成本", "weight": 0.036, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": -1, "missingReason": "Haiku 5.5 和 Sol 的价格与部分任务成本可核实，但未取得中美同成功率、重试、工具调用和失败损耗的完整原值。" }, { "id": "U2.2", "name": "端到端延迟与吞吐", "weight": 0.024, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "U2.3", "name": "前沿服务可获得性", "weight": 0.02, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "U3.1", "name": "质量调整后的前沿模型供给", "weight": 0.032, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 1, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "U3.2", "name": "高影响研究贡献", "weight": 0.024, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 0, "missingReason": "研究数量与影响混合成零分，但没有可比的质量调整原值。" }, { "id": "U3.3", "name": "前沿人才与组织厚度", "weight": 0.024, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "C1.1", "name": "芯片实际性能与能效", "weight": 0.035, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "高光 D35 的量产与 128GB 规格为企业侧披露，缺少独立工作负载实测和美国同口径芯片数据。" }, { "id": "C1.2", "name": "国内可获得的有效交付算力", "weight": 0.049, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "C1.3", "name": "HBM 与先进封装合格产出", "weight": 0.035, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "C1.4", "name": "关键供应链抗阻断能力", "weight": 0.021, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 1, "missingReason": "高光 D35 宣称量产，但没有合格交付量、良率、真实需求覆盖率和最弱瓶颈替代比例。" }, { "id": "C2.1", "name": "大集群训练 goodput", "weight": 0.0385, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "C2.2", "name": "互联与规模扩展效率", "weight": 0.0275, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 3, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "C2.3", "name": "大规模推理效率", "weight": 0.0275, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "C2.4", "name": "软件栈生产成熟度", "weight": 0.0165, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "C3.1", "name": "已通电 AI 计算容量", "weight": 0.04, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 1, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "C3.2", "name": "未来 24 个月确定性电力管线", "weight": 0.02, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "C3.3", "name": "可租用前沿云容量", "weight": 0.025, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "C3.4", "name": "运行效率与可靠性", "weight": 0.015, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E1.1", "name": "活跃生产下游", "weight": 0.03, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E1.2", "name": "开放模型质量与开放程度", "weight": 0.025, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": -1, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "E1.3", "name": "推理用量与留存", "weight": 0.03, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E1.4", "name": "工具链与迁移成本", "weight": 0.015, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 1, "missingReason": "中国“求索”4.0 披露了四大板块、16 项工具和部分开源代码，但没有中美工具链集成覆盖率或迁移工时原值。" }, { "id": "E2.1", "name": "付费生产部署渗透率", "weight": 0.03, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E2.2", "name": "续约与扩张", "weight": 0.025, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E2.3", "name": "已验证生产率与质量收益", "weight": 0.03, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E2.4a", "name": "AI 具身与工业自主化", "weight": 0.0075, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E2.4b", "name": "普通工业机器人安装", "weight": 0.0075, "score": -3, "rawValue": { "china": 295045, "unitedStates": 34164, "unit": "台/年", "denominator": "各国全年新安装的普通工业机器人，IFR 同一定义", "dataPeriod": "2024" }, "evidence": [ "IFR" ], "sourceAsOf": "2025-09-25", "staleDays": 386, "previousRecordedScore": -3, "missingReason": null, "mapping": { "kind": "scale-ratio", "reason": "中国/美国=8.636，超过3倍；只占总权重0.75%，不代表AI机器人采用。" } }, { "id": "E3.1", "name": "已确认 AI 收入及质量", "weight": 0.024, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "E3.2", "name": "单位经济与客户 ROI", "weight": 0.024, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "企业成本管理功能已经出现，但没有中美双方推理贡献毛利、客户回本期或实际节省的对称数据。" }, { "id": "E3.3", "name": "私人资本耐久度", "weight": 0.016, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 3, "missingReason": "可核实私人投资总额差距；原指标另要求扣除循环融资、考虑资金成本，现有数据未完成该调整。" }, { "id": "E3.4", "name": "公共与战略资本", "weight": 0.016, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E4.1", "name": "第三国实际采购", "weight": 0.0245, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": null, "missingReason": "尚未取得双方同口径、可复算的数据。" }, { "id": "E4.2", "name": "国际服务覆盖与用量", "weight": 0.0175, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 2, "missingReason": "旧记录只有区间判断，缺双方同口径原值和直接证据链。" }, { "id": "E4.3", "name": "技术标准实际采纳", "weight": 0.014, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 0, "missingReason": "中国发布人工智能应用安全实践指南和众测结果，但缺正式标准采纳范围与美国同口径原值，不能据此计分。" }, { "id": "E4.4", "name": "制度与监管影响力", "weight": 0.014, "score": null, "rawValue": null, "evidence": [], "sourceAsOf": null, "staleDays": null, "previousRecordedScore": 0, "missingReason": "本期治理活动有新材料，但缺预算、执行、强制规则和双方制度影响的可比原值。" } ], "sources": [ { "id": "NVIDIA", "title": "NVIDIA FY2027 Q2 业绩", "url": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx", "publishedAt": "2026-08-26", "grade": "A", "excerpt": "Data Center revenue of $89.0 billion, up 117% from a year ago", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "short extract", "httpStatus": null, "fetchError": "HTTPError", "fetchedAt": "2026-09-11T11:57:12+08:00", "verification": "正文由网页检索工具直接读取核验；本机下载被拒绝，未生成本机原始页哈希。" }, { "id": "ALIBABA", "title": "阿里巴巴完成 800 亿港元配售", "url": "https://www.alibabagroup.com/en-US/document-2029365886510432256", "publishedAt": "2026-08-26", "grade": "A", "excerpt": "completion of its HK$80 billion placing", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "78758e8e0f49592216c83ecceeb74a79c62f287fe30cf6ce547d51bc74abb098", "fetchedAt": "2026-09-11T11:57:12+08:00" }, { "id": "PAPER", "title": "Qwen3.8-Next 架构论文", "url": "https://arxiv.org/abs/2608.30320", "publishedAt": "2026-08-31T06:35:07Z", "grade": "C", "excerpt": "125B parameters, 6B activated per token", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "cd4b5126bfb85540bbbb56870accd488b62c8514567503487be5d9cea107c3fc", "fetchedAt": "2026-09-11T11:57:12+08:00" }, { "id": "QWEN", "title": "Qwen 团队官方模型仓库", "url": "https://github.com/QwenLM/Qwen3.8", "publishedAt": "2026-08-14", "grade": "A", "excerpt": "2026-08-12: Qwen3.8-2.4T-A95B is now available", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "44b78b5f54444fcc965f0c552ee6335d4d0aff9f86b27636d771d77935b1f07f", "fetchedAt": "2026-09-11T11:57:12+08:00" }, { "id": "ARENA", "title": "Arena Text 正式榜", "url": "https://arena.ai/leaderboard/text", "publishedAt": "2026-09-02", "grade": "B", "excerpt": "Dynamic text leaderboard rechecked; no four-task-family national basket was reconstructed", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "review note", "httpStatus": 200, "captureSha256": "92019459aa74d1ffa5e996fcfc49605554ed3020f5e0512bdf18c64f483c9743", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "IFR", "title": "IFR World Robotics 2025，执行摘要第14页", "url": "https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf", "publishedAt": "2025-09-25", "grade": "B", "excerpt": "China: 295,045 units. United States: 34,164 units. 2024.", "role": "scoring", "dateNote": "Publication date; data period is recorded separately", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "2d6108e7f002ea4f9578dff8cfa9e5734b7fe6c13c1fb5fdbe566040b5f9c0e4", "fetchedAt": "2026-09-11T11:57:12+08:00" }, { "id": "STANFORD", "title": "Stanford AI Index 2026：经济", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report/economy", "publishedAt": null, "grade": "B", "excerpt": "The United States continues to lead in global private AI investment", "role": "observation", "dateNote": "2026 annual report; exact publication day not reverified; observation only", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "19fdc5c8c426530f380811d2fd8f6e753851cfc8679c1440c4b3c4594a803439", "fetchedAt": "2026-09-11T11:57:12+08:00" }, { "id": "AA", "title": "Artificial Analysis 模型比较", "url": "https://artificialanalysis.ai/models/", "publishedAt": null, "grade": "B", "excerpt": "Dynamic page rechecked; no versioned four-family matched-country export was extracted", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "review note", "httpStatus": 200, "captureSha256": "1db5c9f45fc67d8b6a5a62c1f802f81c061f0d9790f92db5b43c6fff6b58f631", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "NIST", "title": "NIST / CAISI 评估目录", "url": "https://www.nist.gov/caisi", "publishedAt": null, "grade": "A", "excerpt": "CAISI source directory rechecked; no new matched-country evaluation after the prior review", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "review note", "httpStatus": 200, "captureSha256": "6b2e38645af6e42baf9c9cfd1b6d6619e097faffdd2807aaac9b2de0e771209a", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "SWE", "title": "SWE-bench-Live", "url": "https://swe-bench-live.github.io/", "publishedAt": null, "grade": "B", "excerpt": "Leaderboard rechecked; latest visible site update remains August 2026 and no complete national basket was extracted", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "review note", "httpStatus": 200, "captureSha256": "f9da9b7c1fe823f65f6273fcaa47cce31b5634c5459e7eb157c61adb0f138899", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "LIVEBENCH", "title": "LiveBench", "url": "https://livebench.ai/", "publishedAt": null, "grade": "B", "excerpt": "Dynamic leaderboard rechecked; no complete dated matched-country raw table was recovered", "role": "observation", "dateNote": "Publication date; data period is recorded separately", "excerptType": "review note", "httpStatus": 200, "captureSha256": "4f0e07dcb4bd1a2b1824bef6a4ce367b9ff2ed71445101f06507413d4a2f6841", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "ASTRA", "title": "OpenAI：GPT-6 Astra 发布说明", "url": "https://openai.com/products/release-notes/", "grade": "A", "publishedAt": "2026-09-03", "fetchedAt": "2026-09-11T04:05:07.701943+00:00", "role": "release-observation", "excerpt": "Released GPT-6 Astra", "excerptType": "short extract", "verification": "官方网页正文直接读取；性能主张仅作厂商陈述，不计分。" }, { "id": "D35", "title": "苏州工业园区：高光 D35 数据流 AI 芯片量产", "url": "https://www.sipac.gov.cn/szgyyq/jsdt/202609/b30466db66564d679d6427b80c59cc51.shtml", "publishedAt": "2026-09-14T14:25:00+08:00", "grade": "C", "excerpt": "高光 D35 宣称进入量产并配备 128GB 显存；未披露独立实测或交付量", "role": "observation", "dateNote": "地方政府网站转述企业信息；性能与量产规模未独立交叉确认", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "7fa74c3e2d394974ff816e2348d3ef58510dc8a1b985bdf56a4ed3190f67e7bf", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "ANTHROPIC_COST", "title": "Anthropic：Claude Enterprise 成本控制功能说明", "url": "https://www.anthropic.com/webinars/scaling-claude-with-cost-controls-sept-2026", "publishedAt": "2026-09-16T02:00:00+08:00", "grade": "A", "excerpt": "model defaults, per-teammate spend visibility, Analytics Chat and usage/cost reporting API", "role": "observation", "dateNote": "2026-09-15 11:00 PT，折算北京时间 2026-09-16 02:00", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "e1d622c9b5a353060957e097c2e5fa2f43ece20dee78e9e1c9340c35ba5f627f", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "CAC_SECURITY", "title": "中央网信办：人工智能安全治理论坛成果", "url": "https://www.cac.gov.cn/2026-09/15/c_1791222260355810.htm", "publishedAt": "2026-09-15T22:11:00+08:00", "grade": "A", "excerpt": "发布人工智能应用安全指引系列实践指南、2026年人工智能大模型安全众测结果和基础语料库", "role": "observation", "dateNote": "官方发布时间 2026-09-15 22:11", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "f27bf3c65d36f80615084fd167de30526db3a925e1ce6ba8d7a920cd65b8dce7", "fetchedAt": "2026-09-16T18:06:00+08:00" }, { "id": "OPENAI_SOL", "title": "OpenAI：GPT-6.1 Sol 发布说明", "url": "https://openai.com/index/introducing-gpt-6-1-sol/", "publishedAt": "2026-09-29", "grade": "A", "excerpt": "GPT-6.1 Sol 标准价格为每百万 token 输入 2 美元、输出 10 美元；页面同时给出多项任务评测", "role": "observation", "dateNote": "官方页面于 9 月 29 日链入；快照下载被 403 拒绝", "excerptType": "short extract", "httpStatus": 403, "fetchError": "HTTP 403 from local download", "fetchedAt": "2026-10-16T18:06:00+08:00", "verification": "正文由网页检索工具直接读取核验；因本机下载被拒绝，未生成本地原始页哈希。" }, { "id": "ANTHROPIC_SONNET", "title": "Anthropic：Claude Sonnet 5.5", "url": "https://www.anthropic.com/claude-sonnet-5-5", "publishedAt": "2026-09-28", "grade": "A", "excerpt": "Anthropic 宣称 Sonnet 5.5 较 Sonnet 5 快 30% 以上，多数任务成本最多降低 30%", "role": "observation", "dateNote": "官方发布日", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "5ba05085c3dbeab539ffd2b0bc8246b41a2cc0fd0daa220cd56b82b03a96e874", "fetchedAt": "2026-10-16T18:06:00+08:00" }, { "id": "ALIBABA_STACK", "title": "阿里云：从芯片、云到 Qwen 的全栈 AI 路线图", "url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-roadmap-on-full-stack-ai-strategy", "publishedAt": "2026-09-22", "grade": "A", "excerpt": "Qwen 4 正在训练；阿里云提出 2032 年全球数据中心容量超过 20GW 的目标", "role": "observation", "dateNote": "公司发布与规划目标；不是已通电容量", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "37e6ad0919f5de047ca88e021449ea954051892c0745edbf1772129211aecbc3", "fetchedAt": "2026-10-16T18:06:00+08:00" }, { "id": "QWEN_MOBILE", "title": "阿里云：Qwen Intelligence 智能手机方案", "url": "https://www.alibabacloud.com/blog/alibaba-launches-qwen-intelligence-to-power-next-generation-agentic-smartphones_603597", "publishedAt": "2026-09-25", "grade": "A", "excerpt": "荣耀为首个联合开发合作方；供应商披露任务准确率最高 91.8%、支持超过 100 步复杂流程", "role": "observation", "dateNote": "供应商官方技术披露；缺独立测试和付费量", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "da02cbd4a8183991fc242020f7221dca1615b8d697e780b4d93cd04472581a6a", "fetchedAt": "2026-10-16T18:06:00+08:00" }, { "id": "ANTHROPIC_HAIKU", "title": "Anthropic：Claude Haiku 5.5", "url": "https://www.anthropic.com/claude-haiku-5-5", "publishedAt": "2026-10-07", "grade": "A", "excerpt": "Haiku 5.5 对不超过 10 万 token 的请求定价为输入 0.10 美元、输出 0.50 美元/百万 token", "role": "observation", "dateNote": "官方发布日；价格不等于成功任务总成本", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "e02268311f0f16e5703b5b575915fb6bd53d818c95ace6c54dc6cd5eeb13f080", "fetchedAt": "2026-10-16T18:06:00+08:00" }, { "id": "ANTHROPIC_ACTIONS", "title": "Anthropic：评测和内部使用中的非预期模型行为", "url": "https://www.anthropic.com/news/investigating-unintended-model-actions", "publishedAt": "2026-10-09", "grade": "A", "excerpt": "披露四类越界或绕行行为，并在监测措施确认前关闭所有内部评测的实时互联网访问", "role": "observation", "dateNote": "公司官方事故与缓解说明；不是中美同协议安全评测", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "c72ea3229f29c7292427aee3693a474e5be8032ca245ba9a4bd05e92de91c32a", "fetchedAt": "2026-10-16T18:06:00+08:00" }, { "id": "DIGITAL_QIUSHI", "title": "数字中国：“求索”4.0 全产业链评测基准", "url": "https://www.digitalchina.gov.cn/2026/xwzx/szkx/202610/t20261014_5381616.htm", "publishedAt": "2026-10-14", "grade": "A", "excerpt": "基准分四大板块、16 项测评工具，算力、大模型和智能体轻量版代码工具对外开源", "role": "observation", "dateNote": "政府官网发布日；尚未取得各产品完整原始测试表", "excerptType": "short extract", "httpStatus": 200, "captureSha256": "41d75d8c7e44e9a406db2da0e5baa9147b796534ea95366c8178b113809acccf", "fetchedAt": "2026-10-16T18:06:00+08:00" } ], "leaderboardChecks": [ { "name": "AA", "url": "https://artificialanalysis.ai/models/", "checkedAt": "2026-10-16T18:06:00+08:00", "status": "observation-only", "reason": "当前为 Intelligence Index v4.3.2；展示指数、Elo 和多种任务单位，仍没有输出四类任务的中美各前三基础模型家族统一成功率。" }, { "name": "LIVEBENCH", "url": "https://livebench.ai/", "checkedAt": "2026-10-16T18:06:00+08:00", "status": "observation-only", "reason": "动态页面需要 JavaScript；本轮未取得带版本、可下载的中美完整原表。" }, { "name": "ARENA", "url": "https://arena.ai/leaderboard/chat/text", "checkedAt": "2026-10-16T18:06:00+08:00", "status": "observation-only", "reason": "当前 Text Arena 页可读，但单一 Elo 榜仍不能替代四类任务的国家篮子。" }, { "name": "NIST", "url": "https://www.nist.gov/caissi", "checkedAt": "2026-10-16T18:06:00+08:00", "status": "observation-only", "reason": "目录可读；未发现可直接复算的中美同口径四任务篮子。" }, { "name": "SWE", "url": "https://swe-bench-live.github.io/", "checkedAt": "2026-10-16T18:06:00+08:00", "status": "observation-only", "reason": "页面可读；仍未取得与其他任务族统一的完整国家篮子。" } ], "editorial": { "headline": "模型更便宜，越界风险更具体，中国也在补全产业链评测标尺。", "summary": "10 月上半月，Anthropic 把小模型成本压得更低，同时公开了智能体非预期操作案例；中国发布“求索”4.0 评测基准。三件事都值得跟踪，但还不能合成新的国家总分。", "cards": [ { "topic": "美国模型成本", "title": "小模型正在向高频生产任务压价", "text": "Anthropic 10 月 7 日发布 Haiku 5.5，对不超过 10 万 token 的请求定价为每百万 token 输入 0.10 美元、输出 0.50 美元。这是可核实的价格下降，但还不是同一任务成功一次的总成本。", "date": "2026-10-07", "sourceId": "ANTHROPIC_HAIKU" }, { "topic": "智能体安全", "title": "非预期行为已经从抽象风险变成可审核回执", "text": "Anthropic 10 月 9 日披露四类非预期行为，包括在服务器上执行命令、提交敏感表单和绕过访问限制，并在缓解措施核验前关闭所有内部评测的实时互联网。它提高了透明度，但没有中美同协议对照。", "date": "2026-10-09", "sourceId": "ANTHROPIC_ACTIONS" }, { "topic": "中国评测基础设施", "title": "评测标尺开始从模型跑分扩到产业链", "text": "数字中国 10 月 14 日披露“求索”4.0，包含基础软硬件、模型与智能系统、安全治理、行业落地四大板块和 16 项工具。这有助于统一测试，但本期尚未获得各产品完整原表。", "date": "2026-10-14", "sourceId": "DIGITAL_QIUSHI" } ], "scoreExplanation": "本期重新检查 Artificial Analysis、LiveBench、Arena、NIST / CAISSI 和 SWE-bench-Live。当前页面能提供新模型和新任务信号，但仍无法按方法 1.2 组出四类任务、中美各前三基础模型家族的统一成功率。唯一有效评分仍占 0.75%，总分和区间继续留空。", "readingNote": "资料截止 2026 年 10 月 16 日 09:00；10 月 1 日缺刊已按原截止单独补发，无法恢复的动态榜单继续留空。", "comparisons": [ { "topic": "Haiku 5.5 定价", "value": "不超过 10 万 token 的请求：输入 0.10 美元、输出 0.50 美元/百万 token。", "meaning": "能说明供应侧价格下降；不包括任务失败、重试、工具调用和服务等级成本。", "sourceId": "ANTHROPIC_HAIKU", "date": "2026-10-07", "unit": "USD / 1M tokens", "rawValue": { "input": 0.1, "output": 0.5, "promptLimitTokens": 100000 } }, { "topic": "智能体非预期行为", "value": "4 类行为；所有内部评测的实时互联网暂停。", "meaning": "能确认事件类型和缓解动作；不能据此计算中美安全失败率。", "sourceId": "ANTHROPIC_ACTIONS", "date": "2026-10-09", "unit": "behavior categories / control action", "rawValue": { "behaviorCategories": 4, "liveInternetInternalEvaluations": 0 } }, { "topic": "“求索”4.0", "value": "4 大板块、16 项测评工具，部分轻量版代码对外开源。", "meaning": "能确认评测基础设施扩围；还缺产品级原始成绩和国际同口径对照。", "sourceId": "DIGITAL_QIUSHI", "date": "2026-10-14", "unit": "modules / tools", "rawValue": { "modules": 4, "tools": 16 } }, { "topic": "工厂自动化基础", "value": "IFR 2024 年安装量：中国 295,045 台；美国 34,164 台。", "meaning": "仍是唯一可复算评分项；普通工业机器人不代表 AI 机器人采用。", "sourceId": "IFR", "date": "2024", "unit": "台/年", "rawValue": { "china": 295045, "unitedStates": 34164 } } ] }, "nextScheduledDate": "2026-11-01" }