{
  "schemaVersion": "1.0",
  "id": "fortune-telling",
  "title": "命理年度详解",
  "shortTitle": "AI 算命模型横评",
  "description": "让不同模型解读同一份匿名年度命理样本，比较中文表达、命理依据、现实信号、行动建议与结构化输出稳定性。",
  "capturedAt": "2026-06-01",
  "publishedAt": "2026-07-22",
  "sourceName": "Kline-star AI Model Evaluations",
  "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/compare-full-model-set-2026.md",
  "repositoryUrl": "https://github.com/k2009/Kline-star",
  "machineDataPath": "/data/chinese-evaluations.json",
  "fixture": {
    "id": "anonymous-2026-high-opportunity-high-risk",
    "targetYear": 2026,
    "sampleCount": 1,
    "description": "同一份匿名固定样本：事业与财富得分较高、健康得分较低，用于检验模型能否解释高机会与高风险并存的年度状态。"
  },
  "runs": [
    {
      "id": "2026-05-01T17-27-31-325Z",
      "provider": "TencentMaaS",
      "createdAt": "2026-05-01T17:35:24.345Z",
      "temperature": 0.2,
      "maxTokens": 2400,
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/README.md"
    },
    {
      "id": "2026-06-01T02-00-33-587Z",
      "provider": "Xiaomi MiMo Token Plan",
      "createdAt": "2026-06-01T02:01:35.380Z",
      "temperature": 0.2,
      "maxTokens": 4000,
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-06-01T02-00-33-587Z/README.md"
    }
  ],
  "criteria": [
    "命理依据是否解释大运、流年、藏干、十神与岁运并临，而不是只堆术语",
    "是否把事业、财富、关系与健康的强弱翻译成具体体感和现实信号",
    "中文判词、叙事结构与行动建议是否清晰、具体且适合产品阅读",
    "JSON 是否可解析，必需字段是否完整，输出是否被截断"
  ],
  "limitations": [
    "这是单一匿名固定样本，不代表模型在所有命理任务上的平均水平。",
    "内容结论来自该批次人工阅读记录，没有统一数值总分；本站不重新评分。",
    "小米模型来自后续独立批次，最大输出长度不同，因此延迟不宜与主批次直接排名。",
    "本专题评估的是中文产品输出质量与结构稳定性，不验证命理预测是否具有科学有效性。"
  ],
  "models": [
    {
      "id": "glm-5v-turbo",
      "displayName": "GLM-5V-Turbo",
      "modelId": "glm-5v-turbo",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 88435,
      "completionTokens": 2400,
      "note": "原横评认为其文本厚、情景化强，行动提醒清晰，中文可读性最突出。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/glm-5v-turbo.md",
      "latencyComparable": true
    },
    {
      "id": "minimax-m2-5",
      "displayName": "MiniMax-M2.5",
      "modelId": "minimax-m2.5",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 27210,
      "completionTokens": 2045,
      "note": "现实落地逻辑强，风险与执行建议的对比鲜明，实操价值突出。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/minimax-m2-5.md",
      "latencyComparable": true
    },
    {
      "id": "deepseek-v4-pro",
      "displayName": "DeepSeek-V4-Pro",
      "modelId": "deepseek-v4-pro",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 43705,
      "completionTokens": 1616,
      "note": "命理锚点明确、结构完整、建议可执行，整体表达稳定。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/deepseek-v4-pro.md",
      "latencyComparable": true
    },
    {
      "id": "deepseek-v4-flash",
      "displayName": "DeepSeek-V4-Flash",
      "modelId": "deepseek-v4-flash",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 21508,
      "completionTokens": 1564,
      "note": "开头抓人，逻辑链完整，语言利落，适合直接面向中文用户。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/deepseek-v4-flash.md",
      "latencyComparable": true
    },
    {
      "id": "deepseek-v3-2",
      "displayName": "DeepSeek-V3.2",
      "modelId": "deepseek-v3.2",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 39159,
      "completionTokens": 1309,
      "note": "结构稳定、语义完整，风格更接近传统命理叙事。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/deepseek-v3-2.md",
      "latencyComparable": true
    },
    {
      "id": "hy3-preview",
      "displayName": "Hy3 Preview",
      "modelId": "hy3-preview",
      "providerBatch": "TencentMaaS",
      "result": "ok",
      "recommendation": "recommended",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 43719,
      "completionTokens": 977,
      "note": "篇幅较短，但主次清楚、结构紧凑，适合作为轻量版本。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/hy3-preview.md",
      "latencyComparable": true
    },
    {
      "id": "minimax-m2-7",
      "displayName": "MiniMax-M2.7",
      "modelId": "minimax-m2.7",
      "providerBatch": "TencentMaaS",
      "result": "bad-shape",
      "recommendation": "conditional",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 85844,
      "completionTokens": 1872,
      "note": "内容质量接近可用，但额外包裹 Markdown 代码围栏，导致结构化解析失败。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/minimax-m2-7.md",
      "latencyComparable": true
    },
    {
      "id": "glm-5",
      "displayName": "GLM-5",
      "modelId": "glm-5",
      "providerBatch": "TencentMaaS",
      "result": "bad-shape",
      "recommendation": "conditional",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 42517,
      "completionTokens": 2400,
      "note": "原文可读，但 JSON 换行或转义异常且触及输出上限，需要修复后再用。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/glm-5.md",
      "latencyComparable": true
    },
    {
      "id": "mimo-v2-5-pro",
      "displayName": "Xiaomi MiMo V2.5 Pro",
      "modelId": "mimo-v2.5-pro",
      "providerBatch": "Xiaomi MiMo Token Plan",
      "result": "ok",
      "recommendation": "conditional",
      "strictJson": true,
      "completeShape": true,
      "elapsedMs": 61783,
      "completionTokens": 1258,
      "note": "第二次调用通过严格 JSON；表达完整但偏模板化，原记录列为二级候选。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-06-01T02-00-33-587Z/outputs/mimo-v2-5-pro.md",
      "latencyComparable": false
    },
    {
      "id": "glm-5-1",
      "displayName": "GLM-5.1",
      "modelId": "glm-5.1",
      "providerBatch": "TencentMaaS",
      "result": "bad-shape",
      "recommendation": "conditional",
      "strictJson": true,
      "completeShape": false,
      "elapsedMs": 52150,
      "completionTokens": 2400,
      "note": "可读部分语义质量好，但输出长度截断，必需字段不完整。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/glm-5-1.md",
      "latencyComparable": true
    },
    {
      "id": "deepseek-v3-0324",
      "displayName": "DeepSeek-V3.0324",
      "modelId": "deepseek-v3-0324",
      "providerBatch": "TencentMaaS",
      "result": "bad-shape",
      "recommendation": "conditional",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 26980,
      "completionTokens": 794,
      "note": "文本内容仍可读，但额外包裹 Markdown 代码围栏，导致 JSON 解析失败。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/deepseek-v3-0324.md",
      "latencyComparable": true
    },
    {
      "id": "kimi-k2-5",
      "displayName": "Kimi K2.5",
      "modelId": "kimi-k2.5",
      "providerBatch": "TencentMaaS",
      "result": "failed",
      "recommendation": "not-comparable",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 546,
      "completionTokens": null,
      "note": "调用参数不兼容：该模型只接受 temperature=1，因此本批次没有可比较输出。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/kimi-k2-5.md",
      "latencyComparable": true
    },
    {
      "id": "kimi-k2-6",
      "displayName": "Kimi K2.6",
      "modelId": "kimi-k2.6",
      "providerBatch": "TencentMaaS",
      "result": "failed",
      "recommendation": "not-comparable",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 588,
      "completionTokens": null,
      "note": "调用参数不兼容：该模型只接受 temperature=1，因此本批次没有可比较输出。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/kimi-k2-6.md",
      "latencyComparable": true
    },
    {
      "id": "deepseek-v3-1",
      "displayName": "DeepSeek-V3.1",
      "modelId": "deepseek-v3.1",
      "providerBatch": "TencentMaaS",
      "result": "failed",
      "recommendation": "not-comparable",
      "strictJson": false,
      "completeShape": false,
      "elapsedMs": 341,
      "completionTokens": null,
      "note": "评测时 API 返回模型不存在，本批次没有可比较输出。",
      "sourceUrl": "https://github.com/k2009/Kline-star/blob/main/docs/ai-model-evals/tencentmaas-yearly-report/2026-05-01T17-27-31-325Z/outputs/deepseek-v3-1.md",
      "latencyComparable": true
    }
  ]
}