{
 "bench_id": "b002-gazou",
 "slug": "gazou-2026-09-27",
 "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": 1,
 "run_date": "2026-09-27",
 "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": 37,
  "excluded": [
   {
    "reason": "our_error",
    "n": 35
   }
  ],
  "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": "2026-09-26",
  "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-2026-09-27/latest.csv"
 },
 "ais": [
  {
   "name": "ChatGPT",
   "plan": "ChatGPT Pro",
   "model_shown": "中程度",
   "checked_at": "2026-09-26 21:21"
  },
  {
   "name": "Gemini",
   "plan": "Google AI Pro",
   "model_shown": "Flash 拡張",
   "checked_at": "2026-09-26 21:21"
  },
  {
   "name": "Claude",
   "plan": "Claude Max",
   "model_shown": "Opus 5.5 中",
   "checked_at": "2026-09-26 21:21"
  },
  {
   "name": "Copilot",
   "plan": "Microsoft 365 Personal",
   "model_shown": null,
   "checked_at": "2026-09-26 21:21"
  }
 ],
 "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・Claudeの100.0%で同率だった",
  "winner": null,
  "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": 12,
   "trials": 12,
   "accuracy": 1.0,
   "accuracy_display": 100.0,
   "seconds_median": 14.4,
   "seconds_min": 14.1,
   "seconds_max": 18.3,
   "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": 6,
    "human_check": 0
   }
  },
  {
   "ai": "Gemini",
   "correct": 10,
   "trials": 12,
   "accuracy": 0.8333,
   "accuracy_display": 83.3,
   "seconds_median": 19.5,
   "seconds_min": 19.4,
   "seconds_max": 36.6,
   "consistency": 1.0,
   "consistency_display": 100.0,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": 0.8182,
   "instruction_follow_display": 81.8,
   "failures": {
    "refused": 1,
    "timeout": 0,
    "limit": 0,
    "our_error": 6,
    "human_check": 0
   }
  },
  {
   "ai": "Claude",
   "correct": 13,
   "trials": 13,
   "accuracy": 1.0,
   "accuracy_display": 100.0,
   "seconds_median": 14.0,
   "seconds_min": 13.9,
   "seconds_max": 15.0,
   "consistency": 1.0,
   "consistency_display": 100.0,
   "source_alive": null,
   "source_alive_display": null,
   "instruction_follow": 0.9231,
   "instruction_follow_display": 92.3,
   "failures": {
    "refused": 0,
    "timeout": 0,
    "limit": 0,
    "our_error": 5,
    "human_check": 0
   }
  },
  {
   "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": 18,
    "human_check": 0
   }
  }
 ],
 "tasks": [
  {
   "task_id": "t1",
   "level": "易",
   "prompt": "添付したレシートの画像から、3つの数を読み取ってください。\n\n決まり\n・答えはちょうど3行。小計: ◯／消費税: ◯／合計: ◯ の順で書く\n・数は半角で書く。円記号とカンマは付けなくてよい。説明は書かない",
   "scorer": "exact_number",
   "answer_key": "（機械採点 exact_number）小計＝1998（誤差±0）／消費税＝159（誤差±0）／合計＝2157（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/receipt.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.2,
       "chars": 26,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_01_t1.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.2,
       "chars": 26,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "ChatGPT: 入力欄が無い（ログイン切れ）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 26.5,
       "chars": 26,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_01_t1.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.4,
       "chars": 26,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "Gemini: 入力欄が無い（ログイン切れ）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 51,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_01_t1.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 81,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": true,
       "outcome": "scored",
       "seconds": 15.0,
       "chars": 51,
       "why": "数値すべて一致：小計 ○（正解1998）／消費税 ○（正解159）／合計 ○（正解2157）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t1_claude_3.txt",
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "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、お釣りの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。お釣り: ◯ 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）お釣り＝843（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/receipt.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.3,
       "chars": 8,
       "why": "数値すべて一致：お釣り ○（正解843）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_02_t2.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.2,
       "chars": 8,
       "why": "数値すべて一致：お釣り ○（正解843）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（ChatGPT: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": false,
       "outcome": "refused",
       "seconds": 22.9,
       "chars": 35,
       "why": "断られた（ポリシー）。不正解として分母に入れる",
       "instruction_follow": null,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_02_t2.png"
      },
      {
       "n": 2,
       "ok": false,
       "outcome": "scored",
       "seconds": 36.6,
       "chars": 33,
       "why": "数値が違う：お釣り ×（正解843・釣 の行が無い）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Gemini: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.1,
       "chars": 35,
       "why": "数値すべて一致：お釣り ○（正解843）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_02_t2.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 59,
       "why": "数値すべて一致：お釣り ○（正解843）",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t2_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "Claude: 入力欄が無い（ログイン切れ）",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t3",
   "level": "中",
   "prompt": "添付したレシートの画像から、買った品物を全部書き出してください。\n\n決まり\n・1品につき1行。行の先頭は ・ にする。品名と金額を書く\n・小計・消費税・合計・お預り・お釣りの行は書かない",
   "scorer": "set_match",
   "answer_key": "（機械採点 set_match）納豆 3パック［納豆＋168］／鶏むね肉 2枚［鶏むね肉＋412］／牛乳 1000ml［牛乳＋322］／ティッシュ 5箱［ティッシュ＋380］／トマト 3個［トマト＋340］／ヨーグルト 400g［ヨーグルト＋230］／うどん 3玉［うどん＋146］　作り足しは0件まで",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/receipt.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.2,
       "chars": 101,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_03_t3.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.3,
       "chars": 101,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（ChatGPT: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.5,
       "chars": 114,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_03_t3.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.5,
       "chars": 114,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": false,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Gemini: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 131,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_03_t3.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 131,
       "why": "7件すべて一致・作り足し0件",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t3_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Claude: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "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で、棒が一番高い月はどれですか。\n\n決まり\n・答えはちょうど1行。◯月 の形で書く\n・数値と説明は書かない",
   "scorer": "contains_all",
   "answer_key": "（機械採点 contains_all）いちばん高い月＝1月",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/chart.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.5,
       "chars": 2,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.3,
       "chars": 2,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（ChatGPT: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 20.1,
       "chars": 2,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.4,
       "chars": 2,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Gemini: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.3,
       "chars": 21,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_04_t4.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.4,
       "chars": 21,
       "why": "1項目すべてあり",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t4_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Claude: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    }
   ]
  },
  {
   "task_id": "t5",
   "level": "難",
   "prompt": "添付した棒グラフの画像から、月ごとの数値を全部読み取ってください。\n\n決まり\n・1か月につき1行。◯月: ◯◯◯ の形で、左から順に書く\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）1月＝9300（誤差±0）／2月＝8700（誤差±0）／3月＝7000（誤差±0）／4月＝5200（誤差±0）／5月＝4100（誤差±0）／6月＝2900（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/chart.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 18.3,
       "chars": 53,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_05_t5.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.1,
       "chars": 53,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（ChatGPT: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 20.0,
       "chars": 59,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_05_t5.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.4,
       "chars": 59,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Gemini: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.7,
       "chars": 85,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_05_t5.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.1,
       "chars": 85,
       "why": "数値すべて一致：1月 ○（正解9300）／2月 ○（正解8700）／3月 ○（正解7000）／4月 ○（正解5200）／5月 ○（正解4100）／6月 ○（正解2900）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t5_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Claude: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "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）から、商品Aの列の数値を全部読み取ってください。\n\n決まり\n・1行につき1つ。月: ◯◯◯ の形で、上から順に書く\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）2025-07＝2300（誤差±0）／2025-08＝2300（誤差±0）／2025-09＝3500（誤差±0）／2025-10＝5800（誤差±0）",
   "materials": [
    "/data/research/bench/gazou-2026-09-27/materials/table_photo.png"
   ],
   "per_ai": [
    {
     "ai": "ChatGPT",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 17.5,
       "chars": 55,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_chatgpt_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/chatgpt_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.1,
       "chars": 55,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_chatgpt_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（ChatGPT: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Gemini",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 25.7,
       "chars": 72,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_gemini_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/gemini_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 19.4,
       "chars": 59,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_gemini_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Gemini: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Claude",
     "trials": [
      {
       "n": 1,
       "ok": true,
       "outcome": "scored",
       "seconds": 14.0,
       "chars": 90,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_claude_1.txt",
       "shot": "/data/research/bench/gazou-2026-09-27/shots/claude_06_t6.png"
      },
      {
       "n": 2,
       "ok": true,
       "outcome": "scored",
       "seconds": 13.9,
       "chars": 90,
       "why": "数値すべて一致：2025-07 ○（正解2300）／2025-08 ○（正解2300）／2025-09 ○（正解3500）／2025-10 ○（正解5800）",
       "instruction_follow": true,
       "answer_url": "/data/research/bench/gazou-2026-09-27/answers/t6_claude_2.txt",
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Claude: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      }
     ]
    },
    {
     "ai": "Copilot",
     "trials": [
      {
       "n": 1,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 2,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "instruction_follow": null,
       "answer_url": null,
       "shot": null
      },
      {
       "n": 3,
       "ok": null,
       "outcome": "our_error",
       "seconds": null,
       "chars": 0,
       "why": "先に止めた（Copilot: 入力欄が無い（ログイン切れ））",
       "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",
  "プランは各AIの有料版（ChatGPT Pro／Google AI Pro／Claude Max／Microsoft 365 Personal）。無料版の結果ではない",
  "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（測定中にログインが切れたため）。表の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-2026-09-27"
}