{
  "schemaVersion": "2.0",
  "meta": {
    "date": "2026-11-19",
    "comparedTo": "2026-11-18",
    "generatedAt": "2026-11-19T15:20:00Z"
  },
  "provenance": {
    "briefSources": [
      "aa_index",
      "aa_models",
      "cursorbench",
      "terminal_bench"
    ],
    "scatterSources": {
      "aa": "aa_index",
      "cursor": "cursorbench"
    },
    "tokenBreakdownSource": "aa_index",
    "tableSources": [
      "aa_index",
      "aa_models",
      "cursorbench",
      "terminal_bench",
      "frontier_code",
      "aa_omniscience"
    ]
  },
  "models": [
    {
      "id": "claude-fable-5-2-max",
      "name": "Claude Fable 5.2",
      "shortLabel": "Fable 5.2 max",
      "effort": "max",
      "vendor": "anthropic",
      "weights": "closed",
      "aaSlug": "claude-fable-5-2",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 8,
        "outputPer1M": 40
      },
      "metrics": {
        "aaIndex": 64.6,
        "aaCostUsd": 26.54,
        "aaReasoningTokensK": 193,
        "aaAnswerTokensK": 81,
        "aaTotalOutputTokensK": 274,
        "aaTimeMin": 41.2,
        "cursorBenchPct": 61.3,
        "cursorCostUsd": 48.11,
        "terminalBenchPct": 65.8,
        "frontierCodePct": null,
        "frontierCodeNote": "xhigh: 50.2%",
        "hallucinationPct": 59.3
      },
      "chartOffset": {
        "aa": {
          "dx": -8,
          "dy": 4,
          "anchor": "end"
        },
        "cursor": {
          "dx": -8,
          "dy": 4,
          "anchor": "end"
        }
      }
    },
    {
      "id": "gpt-6-2-astra-max",
      "name": "GPT-6.2 Astra",
      "shortLabel": "Astra max",
      "effort": "max",
      "vendor": "openai",
      "weights": "closed",
      "aaSlug": "gpt-6-2-astra",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 8,
        "outputPer1M": 40,
        "contextCapK": 272
      },
      "metrics": {
        "aaIndex": 62.1,
        "aaCostUsd": 4,
        "aaReasoningTokensK": 26,
        "aaAnswerTokensK": 15,
        "aaTotalOutputTokensK": 41,
        "aaTimeMin": 13.7,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": 62.1,
        "frontierCodePct": null,
        "frontierCodeNote": "6.0: 44.0%",
        "hallucinationPct": 59.1
      },
      "chartOffset": {
        "aa": {
          "dx": -8,
          "dy": -6,
          "anchor": "end"
        }
      }
    },
    {
      "id": "gemini-4-neon-xhigh",
      "name": "Gemini 4 Neon",
      "shortLabel": "Neon xhigh",
      "effort": "xhigh",
      "vendor": "google",
      "weights": "closed",
      "aaSlug": "gemini-4-neon",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 3,
        "outputPer1M": 15
      },
      "metrics": {
        "aaIndex": 61.4,
        "aaCostUsd": 6.44,
        "aaReasoningTokensK": 81,
        "aaAnswerTokensK": 61,
        "aaTotalOutputTokensK": 142,
        "aaTimeMin": 21.5,
        "cursorBenchPct": 65.4,
        "cursorCostUsd": 10.67,
        "terminalBenchPct": 65.2,
        "terminalBenchNote": "DeepMind eval",
        "frontierCodePct": null,
        "hallucinationPct": 14.4
      },
      "chartOffset": {
        "aa": {
          "dx": 8,
          "dy": -5,
          "anchor": "start"
        },
        "cursor": {
          "dx": 8,
          "dy": -6,
          "anchor": "start"
        }
      }
    },
    {
      "id": "gemini-4-carbon-high",
      "name": "Gemini 4 Carbon",
      "shortLabel": "Carbon high",
      "effort": "high",
      "vendor": "google",
      "weights": "closed",
      "aaSlug": "gemini-4-carbon",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 2,
        "outputPer1M": 10,
        "promo": {
          "regularInputPer1M": 4,
          "regularOutputPer1M": 20,
          "expiresAt": "2026-12-01"
        }
      },
      "metrics": {
        "aaIndex": 60,
        "aaCostUsd": 7.76,
        "aaReasoningTokensK": 96,
        "aaAnswerTokensK": 58,
        "aaTotalOutputTokensK": 154,
        "aaTimeMin": 36.1,
        "cursorBenchPct": 61.5,
        "cursorCostUsd": 8.52,
        "terminalBenchPct": 64.5,
        "frontierCodePct": null,
        "hallucinationPct": 11.4
      },
      "chartOffset": {
        "aa": {
          "dx": 8,
          "dy": 4,
          "anchor": "start"
        },
        "cursor": {
          "dx": 8,
          "dy": -6,
          "anchor": "start"
        }
      }
    },
    {
      "id": "gemini-4-neon-high",
      "name": "Gemini 4 Neon",
      "shortLabel": "Neon high",
      "effort": "high",
      "vendor": "google",
      "weights": "closed",
      "aaSlug": "gemini-4-neon-high",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 3,
        "outputPer1M": 15
      },
      "metrics": {
        "aaIndex": 59.6,
        "aaCostUsd": 3.74,
        "aaReasoningTokensK": 35,
        "aaAnswerTokensK": 27,
        "aaTotalOutputTokensK": 62,
        "aaTimeMin": 14.8,
        "cursorBenchPct": 63.7,
        "cursorCostUsd": 6.54,
        "terminalBenchPct": null,
        "terminalBenchNote": "xhigh 65.2%",
        "frontierCodePct": 46.2,
        "automationBenchPct": 56.3,
        "hallucinationPct": 13.9
      },
      "chartOffset": {
        "aa": {
          "dx": -7,
          "dy": 10,
          "anchor": "end"
        },
        "cursor": {
          "dx": -8,
          "dy": -6,
          "anchor": "end"
        }
      }
    },
    {
      "id": "claude-opus-5-5-max",
      "name": "Claude Opus 5.5",
      "shortLabel": "Opus 5.5 max",
      "effort": "max",
      "vendor": "anthropic",
      "weights": "closed",
      "aaSlug": "claude-opus-5-5",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 4,
        "outputPer1M": 20
      },
      "metrics": {
        "aaIndex": 59.1,
        "aaCostUsd": 10.04,
        "aaReasoningTokensK": 120,
        "aaAnswerTokensK": 53,
        "aaTotalOutputTokensK": 173,
        "aaTimeMin": 24.6,
        "cursorBenchPct": 58.1,
        "cursorCostUsd": 11.88,
        "terminalBenchPct": 65.8,
        "frontierCodePct": 47.3,
        "hallucinationPct": 50.3
      },
      "chartOffset": {
        "aa": {
          "dx": 8,
          "dy": 14,
          "anchor": "start"
        },
        "cursor": {
          "dx": 8,
          "dy": 4,
          "anchor": "start"
        }
      }
    },
    {
      "id": "gpt-6-2-sol-max",
      "name": "GPT-6.2 Sol",
      "shortLabel": "Sol max",
      "effort": "max",
      "vendor": "openai",
      "weights": "closed",
      "aaSlug": "gpt-6-2-sol",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 2,
        "outputPer1M": 10,
        "contextCapK": 272
      },
      "metrics": {
        "aaIndex": 57.3,
        "aaCostUsd": 1.01,
        "aaReasoningTokensK": 23,
        "aaAnswerTokensK": 17,
        "aaTotalOutputTokensK": 40,
        "aaTimeMin": 8.5,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": 61.5,
        "frontierCodePct": 43.8,
        "hallucinationPct": 55.4
      },
      "chartOffset": {
        "aa": {
          "dx": -7,
          "dy": -6,
          "anchor": "end"
        }
      }
    },
    {
      "id": "gemini-4-hydrogen-xhigh",
      "name": "Gemini 4 Hydrogen",
      "shortLabel": "Hydrogen xhigh",
      "effort": "xhigh",
      "vendor": "google",
      "weights": "closed",
      "aaSlug": "gemini-4-hydrogen",
      "showInTokenBreakdown": true,
      "pricing": {
        "inputPer1M": 1.5,
        "outputPer1M": 7.5
      },
      "metrics": {
        "aaIndex": 53.9,
        "aaCostUsd": 3.16,
        "aaReasoningTokensK": 68,
        "aaAnswerTokensK": 42,
        "aaTotalOutputTokensK": 110,
        "aaTimeMin": 11.2,
        "cursorBenchPct": 61.4,
        "cursorCostUsd": 8.09,
        "terminalBenchPct": 56.4,
        "terminalBenchNote": "high",
        "frontierCodePct": null,
        "hallucinationPct": 8.5
      },
      "chartOffset": {
        "aa": {
          "dx": -7,
          "dy": 12,
          "anchor": "end"
        },
        "cursor": {
          "dx": -8,
          "dy": 12,
          "anchor": "end"
        }
      }
    },
    {
      "id": "kimi-k3-1-pro",
      "name": "Kimi K3.1 Pro",
      "shortLabel": "Kimi K3.1 Pro",
      "effort": "default",
      "vendor": "moonshot",
      "weights": "open",
      "license": "Kimi K3.1 Pro",
      "aaSlug": "kimi-k3-1-pro",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 2,
        "outputPer1M": 10
      },
      "metrics": {
        "aaIndex": 46.5,
        "aaCostUsd": 3.68,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": 59.7,
        "frontierCodePct": 25.5,
        "frontierCodeNote": "AA 2.0",
        "hallucinationPct": null
      },
      "chartOffset": {
        "aa": {
          "dx": 7,
          "dy": 4,
          "anchor": "start"
        }
      }
    },
    {
      "id": "qwen4-max",
      "name": "Qwen4 Max",
      "shortLabel": "Qwen4 Max",
      "effort": "default",
      "vendor": "alibaba",
      "weights": "closed",
      "aaSlug": "qwen4-max",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 1.5,
        "outputPer1M": 4.5
      },
      "metrics": {
        "aaIndex": 42.9,
        "aaCostUsd": 3.27,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": null,
        "frontierCodePct": 26.5,
        "frontierCodeNote": "AA 2.0",
        "hallucinationPct": null
      },
      "chartOffset": {
        "aa": {
          "dx": -7,
          "dy": 4,
          "anchor": "end"
        }
      }
    },
    {
      "id": "muse-ember-max",
      "name": "Muse Ember",
      "shortLabel": "Muse Ember max",
      "effort": "max",
      "vendor": "meta",
      "weights": "closed",
      "aaSlug": "muse-ember",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 0.1,
        "outputPer1M": 0.5
      },
      "metrics": {
        "aaIndex": 36,
        "aaCostUsd": 0.14,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": null,
        "terminalBenchNote": "xhigh 30.6%",
        "frontierCodePct": null,
        "hallucinationPct": 17.4
      },
      "chartOffset": {
        "aa": {
          "dx": 7,
          "dy": 4,
          "anchor": "start"
        }
      }
    },
    {
      "id": "grok-4-7-xhigh",
      "name": "Grok 4.7",
      "shortLabel": "Grok 4.7 xhigh",
      "effort": "xhigh",
      "vendor": "xai",
      "weights": "closed",
      "aaSlug": "grok-4-7",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 2,
        "outputPer1M": 6
      },
      "metrics": {
        "aaIndex": 35.5,
        "aaCostUsd": 7.77,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": 48.6,
        "cursorCostUsd": 10.19,
        "terminalBenchPct": 38.8,
        "frontierCodePct": 28.9,
        "hallucinationPct": null
      },
      "chartOffset": {
        "aa": {
          "dx": 7,
          "dy": 4,
          "anchor": "start"
        },
        "cursor": {
          "dx": 7,
          "dy": 4,
          "anchor": "start"
        }
      }
    },
    {
      "id": "deepseek-v4-1-flash-1017-max",
      "name": "DeepSeek V4.1 Flash",
      "shortLabel": "DeepSeek V4.1 Flash",
      "effort": "max",
      "vendor": "deepseek",
      "weights": "open",
      "license": "MIT",
      "aaSlug": "deepseek-v4-1-flash-1017",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 0.3,
        "outputPer1M": 1.2
      },
      "metrics": {
        "aaIndex": 32.6,
        "aaCostUsd": 0.6,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": 38.5,
        "frontierCodePct": null,
        "hallucinationPct": null
      },
      "chartOffset": {
        "aa": {
          "dx": 7,
          "dy": 8,
          "anchor": "start"
        }
      }
    },
    {
      "id": "glm-5-3-max",
      "name": "GLM-5.3",
      "shortLabel": "GLM-5.3 max",
      "effort": "max",
      "vendor": "zhipu",
      "weights": "open",
      "license": "GLM-5.3",
      "aaSlug": "glm-5-3",
      "showInTokenBreakdown": false,
      "pricing": {
        "inputPer1M": 1.4,
        "outputPer1M": 4.4
      },
      "metrics": {
        "aaIndex": 31.7,
        "aaCostUsd": 5.01,
        "aaReasoningTokensK": null,
        "aaAnswerTokensK": null,
        "aaTotalOutputTokensK": null,
        "aaTimeMin": null,
        "cursorBenchPct": null,
        "cursorCostUsd": null,
        "terminalBenchPct": 41.2,
        "frontierCodePct": 21.6,
        "frontierCodeNote": "AA 2.0",
        "hallucinationPct": null
      },
      "chartOffset": {
        "aa": {
          "dx": 7,
          "dy": 4,
          "anchor": "start"
        }
      }
    }
  ],
  "picks": {
    "coding": {
      "upstreamSources": [
        "cursorbench",
        "terminal_bench",
        "frontier_code"
      ],
      "cards": [
        {
          "tier": "default",
          "modelId": "gemini-4-neon-high",
          "primaryMetricKey": "cursorBenchPct",
          "primaryMetricSourceId": "cursorbench",
          "costMetricKey": "cursorCostUsd"
        },
        {
          "tier": "budget",
          "modelId": "gpt-6-2-sol-max",
          "primaryMetricKey": "terminalBenchPct",
          "primaryMetricSourceId": "terminal_bench",
          "costMetricKey": "aaCostUsd"
        },
        {
          "tier": "max",
          "modelId": "gemini-4-neon-xhigh",
          "primaryMetricKey": "cursorBenchPct",
          "primaryMetricSourceId": "cursorbench",
          "costMetricKey": "cursorCostUsd"
        }
      ]
    },
    "office": {
      "upstreamSources": [
        "aa_index",
        "deepmind_eval",
        "aa_omniscience"
      ],
      "cards": [
        {
          "tier": "default",
          "modelId": "gemini-4-neon-high",
          "primaryMetricKey": "automationBenchPct",
          "primaryMetricSourceId": "deepmind_eval",
          "costMetricKey": "aaCostUsd"
        },
        {
          "tier": "budget",
          "modelId": "gemini-4-hydrogen-xhigh",
          "primaryMetricKey": "hallucinationPct",
          "primaryMetricSourceId": "aa_omniscience",
          "costMetricKey": "aaCostUsd"
        },
        {
          "tier": "max",
          "modelId": "gpt-6-2-astra-max",
          "primaryMetricKey": "aaIndex",
          "primaryMetricSourceId": "aa_index",
          "costMetricKey": "aaCostUsd"
        }
      ]
    },
    "general": {
      "upstreamSources": [
        "aa_index",
        "aa_omniscience",
        "aa_models"
      ],
      "cards": [
        {
          "tier": "default",
          "modelId": "gemini-4-neon-high",
          "primaryMetricKey": "aaIndex",
          "primaryMetricSourceId": "aa_index",
          "costMetricKey": "aaCostUsd"
        },
        {
          "tier": "budget",
          "modelId": "gpt-6-2-sol-max",
          "primaryMetricKey": "aaIndex",
          "primaryMetricSourceId": "aa_index",
          "costMetricKey": "aaCostUsd"
        },
        {
          "tier": "max",
          "modelId": "claude-fable-5-2-max",
          "primaryMetricKey": "aaIndex",
          "primaryMetricSourceId": "aa_index",
          "costMetricKey": "aaCostUsd"
        }
      ]
    }
  },
  "editorial": {
    "en": {
      "selectorLabel": "2026-11-19 · 9 Labs Live",
      "headline": "9 Labs Compared on Real Per-Task Spend",
      "seoDescription": "2026-11-19 live snapshot comparing Anthropic, OpenAI, Google, xAI, Meta, Moonshot, Alibaba, DeepSeek, and Zhipu on real per-task USD spend.",
      "brief": [
        {
          "code": "01",
          "text": "Frontier leaders: Neon takes the top two CursorBench spots (65.4% / 63.7%), Astra holds AA spend to $4.00 (62.1), and Fable 5.2 leads AA at 64.6 ($26.54)."
        },
        {
          "code": "02",
          "text": "Open & budget: Kimi K3.1 Pro tops open weights (46.5 AA, 59.7% Terminal-Bench), DeepSeek V4.1 Flash runs at $0.60/task (MIT), and Meta Muse Ember sets a $0.14 floor (36.0)."
        },
        {
          "code": "03",
          "text": "Pricing note: Gemini 4 Carbon's $2/$10 promo ends Nov 30 (reverts to $4/$20 Dec 1; standard AA bill $7.76). Qwen4 Max (42.9, $3.27) is closed API only."
        }
      ],
      "scatterNotes": {
        "cursor": "GPT-6.2, Meta Muse Ember, and Chinese open/API models are not yet listed on CursorBench 4.1."
      },
      "cards": {
        "coding.default": {
          "delta": "CursorBench #2",
          "basis": "Ranks 2nd on CursorBench at $6.54/task, 45% below Opus 5.5 max.",
          "boundary": "Trails Opus on pure terminal tasks (65.8%); peak SWE relies on official CLI."
        },
        "coding.budget": {
          "delta": "Budget CLI",
          "basis": "61.5% on Terminal-Bench and $1.01/task AA spend via 40k output cap.",
          "boundary": "Not yet on CursorBench 4.1; 55.4% hallucination rate; 272K context tier step."
        },
        "coding.max": {
          "delta": "CursorBench #1",
          "basis": "Top multi-file refactor score (65.4%) with 65.2% Terminal-Bench eval.",
          "boundary": "Costs 63% more per task than Neon high ($10.67 vs $6.54) for +1.7 pts."
        },
        "office.default": {
          "delta": "-63% cost vs Opus",
          "basis": "Leads cross-app office automation at 56.3% with 13.9% hallucination rate.",
          "boundary": "Trails GPT-6.2 Astra on complex financial spreadsheet modeling."
        },
        "office.budget": {
          "delta": "8.5% hallucination",
          "basis": "Lowest hallucination rate in the set (8.5%) at $3.16 per AA task.",
          "boundary": "Multi-step reasoning depth (53.9 AA) trails flagship configurations."
        },
        "office.max": {
          "delta": "AA 62.1 at $4.00",
          "basis": "Hits 62.1 AA at $4.00/task and 13.7m average time.",
          "boundary": "Low abstention on unknown facts drives a 59.1% hallucination rate."
        },
        "general.default": {
          "delta": "Balanced · AA 59.6",
          "basis": "Balances 59.6 AA score, $3.74/task spend, and 13.9% hallucination rate.",
          "boundary": "Higher per-task spend than GPT-6.2 Sol ($1.01) or Muse Ember ($0.14)."
        },
        "general.budget": {
          "delta": "AA 57.3 at $1.01",
          "basis": "Finishes AA suite in 8.5m average with compact 40k output tokens.",
          "boundary": "High hallucination (55.4%) on niche facts; pair with web search."
        },
        "general.max": {
          "delta": "AA #1 (64.6)",
          "basis": "Highest live AA Index (64.6) and joint #1 Terminal-Bench (65.8%).",
          "boundary": "Averages $26.54 and 41.2m per AA task due to 274k output tokens."
        }
      }
    },
    "zh": {
      "selectorLabel": "2026-11-19 · 9 家厂商实测",
      "headline": "9 家主流厂商单次实付与多榜实测",
      "seoDescription": "2026-11-19 实测快照：按单次实付美元成本对比 Anthropic、OpenAI、Google、xAI、Meta、Kimi、Qwen、DeepSeek 与智谱的主流模型。",
      "brief": [
        {
          "code": "01",
          "text": "主力格局：Neon 包揽 CursorBench 前二（65.4% / 63.7%），Astra 单次实付压至 $4.00（AA 62.1），Fable 5.2 以 64.6 分居首（$26.54）。"
        },
        {
          "code": "02",
          "text": "开源与低价：Kimi K3.1 Pro 居开源首位（AA 46.5、TB 59.7%），DeepSeek V4.1 Flash（MIT）单次 $0.60，Meta Muse Ember 以 $0.14 锚定成本底线（AA 36.0）。"
        },
        {
          "code": "03",
          "text": "价格变动：Gemini 4 Carbon 的 $2/$10 优惠 11 月 30 日到期，12 月 1 日恢复 $4/$20（AA 标准实付 $7.76）；Qwen4 Max（42.9 分、$3.27）仅提供闭源 API。"
        }
      ],
      "scatterNotes": {
        "cursor": "CursorBench 4.1 暂未收录 GPT-6.2、Meta Muse Ember 及中国开源/API 模型。"
      },
      "cards": {
        "coding.default": {
          "delta": "CursorBench 第 2",
          "basis": "CursorBench 63.7% 排第二，单次 $6.54 比 Opus 5.5 max 便宜 45%。",
          "boundary": "纯终端任务低于 Opus（65.8%）；SWE 高分依赖官方 CLI。"
        },
        "coding.budget": {
          "delta": "低价终端档",
          "basis": "Terminal-Bench 达 61.5%，40k 输出把单次实付压在 $1.01。",
          "boundary": "暂缺 CursorBench 成绩；幻觉率 55.4%；输入超 272K 涨价。"
        },
        "coding.max": {
          "delta": "CursorBench 第 1",
          "basis": "多文件重构第一（65.4%），Terminal-Bench 达 65.2%。",
          "boundary": "比 high 档贵 63%（$10.67 vs $6.54），分数仅高 1.7 个百分点。"
        },
        "office.default": {
          "delta": "比 Opus 省 63%",
          "basis": "跨软件办公 56.3% 排第一，幻觉率仅 13.9%。",
          "boundary": "复杂财务建模不如 GPT-6.2 Astra。"
        },
        "office.budget": {
          "delta": "幻觉率 8.5% 最低",
          "basis": "全场最低幻觉率（8.5%），单次 $3.16，适合长文档核对。",
          "boundary": "多步复杂推理（AA 53.9）弱于主力旗舰。"
        },
        "office.max": {
          "delta": "AA 62.1 / $4.00",
          "basis": "单次仅 $4.00、耗时 13.7m，跑出 62.1 高综合分。",
          "boundary": "对未知事实容易硬猜，幻觉率达 59.1%。"
        },
        "general.default": {
          "delta": "综合均衡 · AA 59.6",
          "basis": "AA 59.6 分、单次 $3.74、幻觉率 13.9%，三项最均衡。",
          "boundary": "单次花费高于 GPT-6.2 Sol（$1.01）和 Muse Ember（$0.14）。"
        },
        "general.budget": {
          "delta": "AA 57.3 / $1.01",
          "basis": "输出压在 40k tokens，平均 8.5m 完成，适合高频问答。",
          "boundary": "冷门知识幻觉率 55.4%，核查事实需配搜索。"
        },
        "general.max": {
          "delta": "AA 64.6 第 1",
          "basis": "AA 综合分第一（64.6），Terminal-Bench（65.8%）并列第一。",
          "boundary": "274k 输出把单次拉高到 $26.54，平均耗时 41.2m。"
        }
      }
    },
    "ja": {
      "selectorLabel": "2026-11-19 · 主要9社実測",
      "headline": "主要9社・14モデルの1回実費とベンチマーク",
      "seoDescription": "2026-11-19 実測スナップショット。Anthropic、OpenAI、Google、xAI、Meta、Kimi、Qwen、DeepSeek、GLM を1回実費で横断比較。",
      "brief": [
        {
          "code": "01",
          "text": "上位の動き：Neon が CursorBench 1・2位（65.4% / 63.7%）、Astra は AA 実費を $4.00（62.1）に抑え、Fable 5.2 は 64.6（$26.54）で首位。"
        },
        {
          "code": "02",
          "text": "オープン・低価格：Kimi K3.1 Pro がオープンモデル首位（AA 46.5・TB 59.7%）、DeepSeek V4.1 Flash（MIT）は1回 $0.60、Meta Muse Ember は $0.14（AA 36.0）で最安。"
        },
        {
          "code": "03",
          "text": "価格の注意点：Gemini 4 Carbon の $2/$10 特価は11月30日終了、12月1日から $4/$20（AA 標準実費 $7.76）。Qwen4 Max（42.9・$3.27）はクローズド API のみ。"
        }
      ],
      "scatterNotes": {
        "cursor": "CursorBench 4.1 には GPT-6.2、Meta Muse Ember、中国系モデルが未掲載。"
      },
      "cards": {
        "coding.default": {
          "delta": "CB 63.7%・2位",
          "basis": "CursorBench 2位を1回 $6.54 で達成。Opus 5.5 max より 45% 安い。",
          "boundary": "純粋な端末操作は Opus（65.8%）に届かず、SWE 上限は公式 CLI が前提。"
        },
        "coding.budget": {
          "delta": "低コスト端末向け",
          "basis": "Terminal-Bench 61.5%。40k 出力に抑え、AA 1回実費は $1.01。",
          "boundary": "CursorBench 未掲載。誤答率 55.4%。272K 超の入力で単価が上がる。"
        },
        "coding.max": {
          "delta": "CB 首位 65.4%",
          "basis": "複数ファイル改修で首位（65.4%）、Terminal-Bench も 65.2%。",
          "boundary": "high 比で1回実費が 63% 増（$10.67 vs $6.54）だが、スコア差は +1.7pt。"
        },
        "office.default": {
          "delta": "Opus 比 -63%",
          "basis": "業務自動化 56.3% で首位、誤答率も 13.9% と低い。",
          "boundary": "複雑な財務モデリングでは GPT-6.2 Astra に及ばない。"
        },
        "office.budget": {
          "delta": "誤答率 8.5%・最小",
          "basis": "全体最小の誤答率（8.5%）、AA 実費 $3.16 で長文チェック向き。",
          "boundary": "多段階の深い推論（AA 53.9）は上位モデルより控えめ。"
        },
        "office.max": {
          "delta": "AA 62.1 / $4.00",
          "basis": "1回 $4.00・平均 13.7m で AA 62.1 の高スコアを記録。",
          "boundary": "不明な事実でも推測しがちで、誤答率は 59.1%。"
        },
        "general.default": {
          "delta": "バランス重視・59.6",
          "basis": "AA 59.6、1回実費 $3.74、誤答率 13.9% と実用バランスが良い。",
          "boundary": "軽めの定常質問なら Sol（$1.01）や Muse Ember（$0.14）が安い。"
        },
        "general.budget": {
          "delta": "AA 57.3 / $1.01",
          "basis": "40k 出力・平均 8.5m で完結し、日常の大量質問で費用を抑えられる。",
          "boundary": "専門外の誤答率が 55.4%。事実確認には検索の併用が必要。"
        },
        "general.max": {
          "delta": "AA 64.6・首位",
          "basis": "AA 総合 64.6 で単独首位、Terminal-Bench（65.8%）も首位タイ。",
          "boundary": "274k 出力により AA 平均実費は $26.54、所要時間も 41.2m と長い。"
        }
      }
    }
  },
  "computed": {
    "paretoAA": [
      {
        "id": "muse-ember-max",
        "name": "Muse Ember (max)",
        "aaIndex": 36,
        "aaCostUsd": 0.14
      },
      {
        "id": "gpt-6-2-sol-max",
        "name": "GPT-6.2 Sol (max)",
        "aaIndex": 57.3,
        "aaCostUsd": 1.01
      },
      {
        "id": "gemini-4-neon-high",
        "name": "Gemini 4 Neon (high)",
        "aaIndex": 59.6,
        "aaCostUsd": 3.74
      },
      {
        "id": "gpt-6-2-astra-max",
        "name": "GPT-6.2 Astra (max)",
        "aaIndex": 62.1,
        "aaCostUsd": 4
      },
      {
        "id": "claude-fable-5-2-max",
        "name": "Claude Fable 5.2 (max)",
        "aaIndex": 64.6,
        "aaCostUsd": 26.54
      }
    ],
    "paretoCursor": [
      {
        "id": "gemini-4-neon-high",
        "name": "Gemini 4 Neon (high)",
        "cursorBenchPct": 63.7,
        "cursorCostUsd": 6.54
      },
      {
        "id": "gemini-4-neon-xhigh",
        "name": "Gemini 4 Neon (xhigh)",
        "cursorBenchPct": 65.4,
        "cursorCostUsd": 10.67
      }
    ],
    "paretoTerminal": [
      {
        "id": "deepseek-v4-1-flash-1017-max",
        "name": "DeepSeek V4.1 Flash (max)",
        "terminalBenchPct": 38.5,
        "aaCostUsd": 0.6
      },
      {
        "id": "gpt-6-2-sol-max",
        "name": "GPT-6.2 Sol (max)",
        "terminalBenchPct": 61.5,
        "aaCostUsd": 1.01
      },
      {
        "id": "gpt-6-2-astra-max",
        "name": "GPT-6.2 Astra (max)",
        "terminalBenchPct": 62.1,
        "aaCostUsd": 4
      },
      {
        "id": "gemini-4-neon-xhigh",
        "name": "Gemini 4 Neon (xhigh)",
        "terminalBenchPct": 65.2,
        "aaCostUsd": 6.44
      },
      {
        "id": "claude-opus-5-5-max",
        "name": "Claude Opus 5.5 (max)",
        "terminalBenchPct": 65.8,
        "aaCostUsd": 10.04
      }
    ]
  },
  "sources": {
    "aa_index": {
      "id": "aa_index",
      "name": "Artificial Analysis Intelligence Index v5.0.1",
      "shortName": "Artificial Analysis v5.0.1",
      "org": "Artificial Analysis",
      "url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index",
      "version": "v5.0.1",
      "accessMethod": "api_free_plus_public"
    },
    "aa_models": {
      "id": "aa_models",
      "name": "Artificial Analysis Model Leaderboard",
      "shortName": "Artificial Analysis",
      "org": "Artificial Analysis",
      "url": "https://artificialanalysis.ai/leaderboards/models/",
      "version": "v5.0.1",
      "accessMethod": "api_free"
    },
    "aa_omniscience": {
      "id": "aa_omniscience",
      "name": "AA-Omniscience v1.1 Hallucination Benchmark",
      "shortName": "AA-Omniscience v1.1",
      "org": "Artificial Analysis",
      "url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index",
      "version": "v1.1",
      "accessMethod": "public_web"
    },
    "cursorbench": {
      "id": "cursorbench",
      "name": "CursorBench 4.1",
      "shortName": "CursorBench 4.1",
      "org": "Anysphere",
      "url": "https://cursor.com/cursorbench",
      "version": "4.1",
      "accessMethod": "public_web"
    },
    "terminal_bench": {
      "id": "terminal_bench",
      "name": "Terminal-Bench 4.1",
      "shortName": "Terminal-Bench 4.1",
      "org": "Snorkel AI / Harbor",
      "url": "https://snorkel.ai/leaderboard/terminal-bench/",
      "version": "4.1",
      "accessMethod": "public_web"
    },
    "frontier_code": {
      "id": "frontier_code",
      "name": "FrontierCode 2.0",
      "shortName": "FrontierCode 2.0",
      "org": "Cognition",
      "url": "https://cognition.com/blog/frontier-code-2.0",
      "version": "2.0",
      "accessMethod": "public_web"
    },
    "deepmind_eval": {
      "id": "deepmind_eval",
      "name": "Google DeepMind Model Evaluation Report",
      "shortName": "DeepMind Eval",
      "org": "Google DeepMind",
      "url": "https://deepmind.google/models/evals-methodology/gemini-4-neon",
      "version": "2026-11",
      "accessMethod": "vendor_report"
    }
  }
}