{
 "bench_id": "b015-gazou",
 "slug": "gazou-2026-11-13",
 "title": "レシート・グラフ・表の画像から数字を読む（4AI・6課題・各3回）",
 "field": "画像・写真の読み取り",
 "field_key": "gazou",
 "round": 2,
 "run_date": "2026-11-13",
 "timezone": "Asia/Tokyo",
 "version": "v1",
 "conductor": "AInformation編集部",
 "funder": "AInformation（自社負担）",
 "sponsor": null,
 "method": {
  "tasks_n": 6,
  "repeats": 3,
  "ais_n": 4,
  "trials_total": 72,
  "trials_valid": 54,
  "excluded": [
   {
    "reason": "human_check",
    "n": 18
   }
  ],
  "new_chat_each_time": true,
  "order_rotated": true,
  "order_by_repeat": {
   "1": [
    "ChatGPT",
    "Gemini",
    "Claude",
    "Copilot"
   ],
   "2": [
    "Gemini",
    "Claude",
    "Copilot",
    "ChatGPT"
   ],
   "3": [
    "Claude",
    "Copilot",
    "ChatGPT",
    "Gemini"
   ]
  },
  "timeout_seconds": 300,
  "scoring": "機械採点のみ（AIに採点させていない）",
  "materials_generated_at": "2026-11-12",
  "materials_note": "自作のレシート画像・棒グラフ画像・少し傾けた表の画像。元の数は作るときに決めてある",
  "counted_outcomes": [
   "scored",
   "refused",
   "timeout"
  ],
  "excluded_outcomes": [
   "limit",
   "our_error",
   "human_check"
  ],
  "answer_cleanup": "画面から答えを読み取るとき、AIによっては読み上げ用の1行や考えている途中の要約が答えと同じ枠に入る。採点の前に「答えの先頭にあって、あとでそっくり同じ行がもう一度出てくる」ところまでを落とす。4AIとも同じ処理をする。落とす前の答えは answers/ にそのまま置いてある",
  "csv_url": "/data/research/bench/gazou-2026-11-13/latest.csv"
 },
 "ais": [
  {
   "name": "ChatGPT",
   "plan": "ChatGPT Plus",
   "paid": true,
   "plan_checked_at": "2026-11-12 19:20",
   "model_shown": "思考量 Medium",
   "checked_at": "2026-11-12 19:55"
  },
  {
   "name": "Gemini",
   "plan": "Google AI Pro",
   "paid": true,
   "plan_checked_at": "2026-11-12 19:20",
   "model_shown": "Flash-Lite",
   "checked_at": "2026-11-12 19:55"
  },
  {
   "name": "Claude",
   "plan": "Claude Ultra",
   "paid": true,
   "plan_checked_at": "2026-11-12 19:20",
   "model_shown": "Opus 4.8  高",
   "checked_at": "2026-11-12 19:55"
  },
  {
   "name": "Copilot",
   "plan": "無料のMicrosoftアカウント",
   "paid": false,
   "plan_checked_at": "2026-11-12 19:20",
   "model_shown": null,
   "checked_at": "2026-11-12 19:55"
  }
 ],
 "summary": {
  "headline": "6課題を各3回ずつ投げたところ、正解率が一番高かったのはChatGPTで15/18試行の83.3%だった（2位はGeminiの66.7%）",
  "winner": "ChatGPT",
  "note": "1回の検証の結果。日や課題が変われば順位は変わる"
 },
 "scores": [
  {
   "ai": "ChatGPT",
   "correct": 15,
   "trials": 18,
   "accuracy": 0.8333,
   "accuracy_display": 83.3,
   "seconds_median": 17.1,
   "seconds_min": 13.9,
   "seconds_max": 26.2,
   "consistency": 1.0,
   "consistency_display": 100.0,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": 1.0,
   "instruction_follow_display": 100.0,
   "failures": {
    "refused": 0,
    "timeout": 0,
    "limit": 0,
    "our_error": 0,
    "human_check": 0
   },
   "below_min": false
  },
  {
   "ai": "Gemini",
   "correct": 12,
   "trials": 18,
   "accuracy": 0.6667,
   "accuracy_display": 66.7,
   "seconds_median": 19.0,
   "seconds_min": 18.9,
   "seconds_max": 43.3,
   "consistency": 1.0,
   "consistency_display": 100.0,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": 0.6667,
   "instruction_follow_display": 66.7,
   "failures": {
    "refused": 0,
    "timeout": 0,
    "limit": 0,
    "our_error": 0,
    "human_check": 0
   },
   "below_min": false
  },
  {
   "ai": "Claude",
   "correct": 10,
   "trials": 18,
   "accuracy": 0.5556,
   "accuracy_display": 55.6,
   "seconds_median": 13.8,
   "seconds_min": 13.8,
   "seconds_max": 16.9,
   "consistency": 0.8889,
   "consistency_display": 88.9,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": 0.1667,
   "instruction_follow_display": 16.7,
   "failures": {
    "refused": 0,
    "timeout": 0,
    "limit": 0,
    "our_error": 0,
    "human_check": 0
   },
   "below_min": false
  },
  {
   "ai": "Copilot",
   "correct": 0,
   "trials": 0,
   "accuracy": null,
   "accuracy_display": null,
   "seconds_median": null,
   "seconds_min": null,
   "seconds_max": null,
   "consistency": null,
   "consistency_display": null,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": null,
   "instruction_follow_display": null,
   "failures": {
    "refused": 0,
    "timeout": 0,
    "limit": 0,
    "our_error": 0,
    "human_check": 18
   },
   "below_min": false
  }
 ],
 "tasks": [
  {
   "task_id": "t1",
   "level": "易",
   "prompt": "添付した棒グラフの画像で、一番高い棒の数値はいくつですか。\n\n決まり\n・答えはちょうど1行。数だけを半角で書く\n・月名と説明は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）一番高い棒の値＝7800（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/chart.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.1,
       "chars": 4,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_01_t1.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 4,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.0,
       "chars": 4,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 18.9,
       "chars": 18,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_01_t1.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 18.9,
       "chars": 5,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 28.1,
       "chars": 5,
       "why": "数値すべて一致：一番高い棒の値 ○（正解7800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 34,
       "why": "数値が違う：一番高い棒の値 ×（正解7800・答え78007800.0）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_01_t1.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 34,
       "why": "数値が違う：一番高い棒の値 ×（正解7800・答え78007800.0）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 36,
       "why": "数値が違う：一番高い棒の値 ×（正解7800・答え78007800.0）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t1_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t2",
   "level": "易",
   "prompt": "添付したレシートの画像から、合計金額を読み取ってください。\n\n決まり\n・答えはちょうど1行。合計: ◯ の形で書く\n・数は半角で書く。説明は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）合計＝2058（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/receipt.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 8,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_02_t2.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 8,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 8,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 25,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_02_t2.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 25,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 25,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 42,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_02_t2.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 42,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 42,
       "why": "数値すべて一致：合計 ○（正解2058）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t2_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t3",
   "level": "中",
   "prompt": "添付した表の画像（少し傾いています）で、月ごとに5商品を足した合計を出してください。\n\n決まり\n・1か月につき1行。◯月: ◯◯◯ の形で、上から順に書く\n・数は半角で書く。計算の途中式は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）2025-07＝61100（誤差±0）／2025-08＝63000（誤差±0）／2025-09＝57700（誤差±0）／2025-10＝50900（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/table_photo.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 20.1,
       "chars": 40,
       "why": "数値すべて一致：2025-07 ○（正解61100）／2025-08 ○（正解63000）／2025-09 ○（正解57700）／2025-10 ○（正解50900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_03_t3.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.1,
       "chars": 63,
       "why": "数値すべて一致：2025-07 ○（正解61100）／2025-08 ○（正解63000）／2025-09 ○（正解57700）／2025-10 ○（正解50900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 63,
       "why": "数値すべて一致：2025-07 ○（正解61100）／2025-08 ○（正解63000）／2025-09 ○（正解57700）／2025-10 ○（正解50900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 22.0,
       "chars": 84,
       "why": "数値が違う：2025-07 ×（正解61100・7月 の行が無い）／2025-08 ×（正解63000・8月 の行が無い）／2025-09 ×（正解57700・9月 の行が無い）／2025-10 ×（正解50900・10月 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_03_t3.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 40.3,
       "chars": 63,
       "why": "数値が違う：2025-07 ×（正解61100・7月 の行が無い）／2025-08 ×（正解63000・8月 の行が無い）／2025-09 ×（正解57700・9月 の行が無い）／2025-10 ×（正解50900・10月 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 28.1,
       "chars": 84,
       "why": "数値が違う：2025-07 ×（正解61100・7月 の行が無い）／2025-08 ×（正解63000・8月 の行が無い）／2025-09 ×（正解57700・9月 の行が無い）／2025-10 ×（正解50900・10月 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 187,
       "why": "数値が違う：2025-07 ×（正解61100・7月 の行が無い）／2025-08 ×（正解63000・8月 の行が無い）／2025-09 ×（正解57700・9月 の行が無い）／2025-10 ×（正解50900・10月 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_03_t3.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 169,
       "why": "数値すべて一致：2025-07 ○（正解61100）／2025-08 ○（正解63000）／2025-09 ○（正解57700）／2025-10 ○（正解50900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 163,
       "why": "数値が違う：2025-07 ×（正解61100・7月 の行が無い）／2025-08 ×（正解63000・8月 の行が無い）／2025-09 ×（正解57700・9月 の行が無い）／2025-10 ×（正解50900・10月 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t3_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t4",
   "level": "中",
   "prompt": "添付した棒グラフの画像の数値を全部足すといくつですか。\n\n決まり\n・答えはちょうど1行。合計: ◯◯◯ の形で書く\n・数は半角で書く。計算の途中式は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）6か月の合計＝43300（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/chart.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 10,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.1,
       "chars": 9,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 9,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 18.9,
       "chars": 9,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 43.3,
       "chars": 22,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 10,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 44,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 44,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 44,
       "why": "数値すべて一致：6か月の合計 ○（正解43300）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t4_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t5",
   "level": "難",
   "prompt": "添付したレシートの画像を読み取り、JSON だけを出力してください。\n\n決まり\n・出力は JSON だけ。前後に説明文を書かない\n・形は {\"items\": [{\"name\": \"品名\", \"price\": 金額}], \"subtotal\": 小計, \"tax\": 消費税, \"total\": 合計} にする\n・数は数値で入れる（文字列にしない・円記号とカンマを入れない）",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）小計＝1906（誤差±0）／消費税＝152（誤差±0）／合計＝2058（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/receipt.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 17.2,
       "chars": 270,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_05_t5.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 17.2,
       "chars": 265,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 17.3,
       "chars": 265,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 473,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_05_t5.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 19.0,
       "chars": 473,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 22.1,
       "chars": 317,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 511,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_05_t5.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 515,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": false,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 318,
       "why": "数値が違う：小計 ×（正解1906・小計 の行が無い）／消費税 ×（正解152・税 の行が無い）／合計 ×（正解2058・合計 の行が無い）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t5_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t6",
   "level": "難",
   "prompt": "添付した表の画像（少し傾いています）の数値を全部足すといくつですか。\n\n決まり\n・答えはちょうど1行。合計: ◯◯◯ の形で書く\n・月の列は足さない。数は半角で書く。計算の途中式は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）表の全部の合計＝232700（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-11-13/materials/table_photo.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 26.2,
       "chars": 10,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/chatgpt_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 23.2,
       "chars": 10,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.1,
       "chars": 10,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_chatgpt_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 21.9,
       "chars": 10,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/gemini_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 22.0,
       "chars": 30,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 18.9,
       "chars": 10,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_gemini_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 140,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-11-13/shots/claude_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 16.9,
       "chars": 126,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.8,
       "chars": 78,
       "why": "数値すべて一致：表の全部の合計 ○（正解232700）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-11-13/answers/t6_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "human_check",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 人間確認が出た（要素 [role=\"dialog\"][aria-label*=\"セキュリティ チェック\"]）。押さずに止めた）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  }
 ],
 "limits": [
  "1日1回の測定である。日や課題が変われば順位は変わる",
  "プランはChatGPT Plus／Google AI Pro／Claude Ultra／無料のMicrosoftアカウント（測る前に画面で確かめた）",
  "Copilotは無料の枠で測っていて、有料プランの結果ではない。Copilotとの差には、使えるモデルや回数の違いも入っている",
  "既定のモデル・既定のモードだけで測っている（Deep Research などは使っていない）",
  "課題は架空の株式会社サンプル商事の自作データ。実在の企業のデータでの結果ではない",
  "機械で採点できる課題だけを扱っている。文章のうまさ・読みやすさは測っていない",
  "ブラウザで人が使うのと同じ画面から測っている。APIの結果ではない",
  "画面に出ているモデル名を読み取れないAIがある回は、その欄を空（null）にしている。読み取れた回はそのまま載せている",
  "この回はCopilotを測れていない（人間確認の画面が出たため）。Copilotの欄は試行0で、ほかのAIと比べられる数字ではない"
 ],
 "license": "CC BY 4.0",
 "attribution": "出典として AInformation調べ と、使ったページのURL・測定日を書いてください。",
 "page_url": "https://ainformation.jp/research/bench/gazou-2026-11-13"
}