{"id":7297,"date":"2026-08-10T09:55:26","date_gmt":"2026-08-10T00:55:26","guid":{"rendered":"https:\/\/techfun.cc\/?p=7297"},"modified":"2026-08-10T09:55:26","modified_gmt":"2026-08-10T00:55:26","slug":"amazon-bedrock-structured-outputs-guide","status":"publish","type":"post","link":"https:\/\/techfun.cc\/generative-ai\/amazon-bedrock-structured-outputs-guide.html","title":{"rendered":"Amazon Bedrock Structured Outputs\u5b9f\u88c5\u30ac\u30a4\u30c9\u2015JSON\u51fa\u529b\u306e\u5909\u9077\u3068\u30d9\u30b9\u30c8\u30d7\u30e9\u30af\u30c6\u30a3\u30b9"},"content":{"rendered":"

\u306f\u3058\u3081\u306b<\/h3>\n

\u751f\u6210AI\u306e\u51fa\u529b\u3092\u30b7\u30b9\u30c6\u30e0\u3067\u6271\u3046\u5834\u5408\u3001\u81ea\u7136\u306a\u6587\u7ae0\u3088\u308a\u3082\u3001\u6c7a\u3081\u3089\u308c\u305f\u9805\u76ee\u3092\u6301\u3064JSON\u304c\u5fc5\u8981\u306b\u306a\u308b\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002\u3057\u304b\u3057\u3001\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u300cJSON\u3067\u8fd4\u3057\u3066\u304f\u3060\u3055\u3044\u300d\u3068\u66f8\u304f\u3060\u3051\u3067\u306f\u3001JSON\u306e\u524d\u5f8c\u306b\u8aac\u660e\u6587\u3084\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u304c\u4ed8\u304f\u3001\u5fc5\u9808\u9805\u76ee\u304c\u6b20\u3051\u308b\u3001\u60f3\u5b9a\u3068\u7570\u306a\u308b\u578b\u304c\u8fd4\u308b\u3068\u3044\u3063\u305f\u554f\u984c\u304c\u6b8b\u308a\u307e\u3059\u3002
\n\u3053\u306e\u554f\u984c\u306b\u5bfe\u3057\u3066\u3001\u3053\u308c\u307e\u3067\u306f\u30d7\u30ed\u30f3\u30d7\u30c8\u306e\u5de5\u592b\u3084Assistant Prefill\u3067\u51fa\u529b\u3092\u8a98\u5c0e\u3057\u3066\u304d\u307e\u3057\u305f\u30022026\u5e742\u6708\u306b\u306fAmazon Bedrock\u3067Structured Outputs\u304c\u4e00\u822c\u63d0\u4f9b\u3055\u308c\u3001JSON Schema\u306b\u6e96\u62e0\u3057\u305f\u51fa\u529b\u3092\u751f\u6210\u6bb5\u968e\u3067\u4fdd\u8a3c\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u308a\u307e\u3057\u305f\uff08AWS\u300cStructured outputs now available in Amazon Bedrock\u300d<\/a>\uff09\u3002<\/p>\n

Structured Outputs\u306e\u57fa\u672c\u6982\u5ff5\u3068\u521d\u6b69\u7684\u306a\u5b9f\u88c5\u306f\u3001\u4ee5\u4e0b\u306e\u8a18\u4e8b\u3067\u7d39\u4ecb\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n

<\/div>
\u751f\u6210AI\u95a2\u9023<\/div>
Structured Outputs\u306e\u57fa\u672c\u3068\u5b9f\u8df5<\/div><\/div><\/a>\n

\u672c\u8a18\u4e8b\u3067\u306f\u3001Amazon Bedrock\u3067JSON\u3092\u5f97\u308b\u65b9\u6cd5\u306e\u5909\u9077\u306b\u3064\u3044\u3066\u3001\u6b21\u306e4\u3064\u306e\u5b9f\u88c5\u4f8b\u3068\u5b9f\u884c\u7d50\u679c\u3092\u6bd4\u8f03\u3057\u307e\u3059\u3002<\/p>\n

    \n
  1. \n \u30d7\u30ed\u30f3\u30d7\u30c8\u3060\u3051\u3067JSON\u3092\u6307\u5b9a\u3059\u308b\n <\/li>\n
  2. \n Assistant Prefill\u3067\u51fa\u529b\u3092\u8a98\u5c0e\u3059\u308b\n <\/li>\n
  3. \n JSON Schema\u51fa\u529b\u5f62\u5f0f\u3067\u6700\u7d42\u56de\u7b54\u306e\u69cb\u9020\u3092\u4fdd\u8a3c\u3059\u308b\n <\/li>\n
  4. \n strict tool use\u3067\u30c4\u30fc\u30eb\u5165\u529b\u306e\u69cb\u9020\u3092\u4fdd\u8a3c\u3059\u308b\n <\/li>\n<\/ol>\n

    3\u30684\u306e\u65b9\u6cd5\u306f\u3001Amazon Bedrock\u306eStructured Outputs\u3092\u69cb\u6210\u3059\u308b2\u3064\u306e\u4ed5\u7d44\u307f\u3067\u3059\u3002\u5358\u7d14\u306a\u512a\u52a3\u3067\u306f\u306a\u304f\u3001\u4fdd\u8a3c\u3059\u308b\u5bfe\u8c61\u304c\u7570\u306a\u308a\u307e\u3059\u30021\u30682\u306f\u65b9\u5f0f\u306e\u5909\u9077\u3092\u7406\u89e3\u3059\u308b\u305f\u3081\u306e\u6bd4\u8f03\u5bfe\u8c61\u3067\u3042\u308a\u3001\u69cb\u9020\u5316\u51fa\u529b\u3092\u5fc5\u8981\u3068\u3059\u308b\u65b0\u898f\u5b9f\u88c5\u306b\u306f\u63a8\u5968\u3057\u307e\u305b\u3093\u3002\u307e\u305aStructured Outputs\u306b\u5bfe\u5fdc\u3059\u308b\u30e2\u30c7\u30eb\u3001API\u3001\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8\u3092\u9078\u5b9a\u3059\u308b\u3053\u3068\u304c\u57fa\u672c\u65b9\u91dd\u3067\u3059\u3002<\/p>\n

    JSON\u51fa\u529b\u3092\u5b89\u5b9a\u3055\u305b\u308b\u65b9\u6cd5\u306e\u5909\u9077<\/h3>\n

    \u30d7\u30ed\u30f3\u30d7\u30c8\u3068Assistant Prefill\u306f\u3001\u30e2\u30c7\u30eb\u306bJSON\u5f62\u5f0f\u3067\u306e\u51fa\u529b\u3092\u4fc3\u3059\u65b9\u6cd5\u3067\u3059\u3002Structured Outputs\u306f\u3001JSON Schema\u306b\u6cbf\u3046\u3088\u3046\u751f\u6210\u7d50\u679c\u305d\u306e\u3082\u306e\u3092\u5236\u7d04\u3057\u307e\u3059\u3002Structured Outputs\u306b\u306f\u3001\u6700\u7d42\u56de\u7b54\u306b\u4f7f\u3046JSON Schema\u51fa\u529b\u5f62\u5f0f\u3068\u3001\u30c4\u30fc\u30eb\u5165\u529b\u306b\u4f7f\u3046strict tool use\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n

    text<\/div>\n
    \r\n\u30d7\u30ed\u30f3\u30d7\u30c8\u3067JSON\u3092\u4f9d\u983c\r\n        \u2193\r\nAssistant Prefill\u3067JSON\u306e\u958b\u59cb\u4f4d\u7f6e\u3092\u8a98\u5c0e\r\n        \u2193\r\nStructured Outputs\u3067\u30b9\u30ad\u30fc\u30de\u6e96\u62e0\u3092\u4fdd\u8a3c\r\n        \u251c\u2500 JSON Schema\u51fa\u529b\u5f62\u5f0f\uff1a\u6700\u7d42\u56de\u7b54\u3092\u4fdd\u8a3c\r\n        \u2514\u2500 strict tool use\uff1a\u30c4\u30fc\u30eb\u5165\u529b\u3092\u4fdd\u8a3c\r\n<\/code><\/pre>\n\n\n\n\n\n\n
    \u65b9\u6cd5<\/th>\nJSON Schema\u306b\u3088\u308b\u4fdd\u8a3c<\/th>\n\u5bfe\u8c61<\/th>\n\u4f4d\u7f6e\u4ed8\u3051<\/th>\n<\/tr>\n
    \u30d7\u30ed\u30f3\u30d7\u30c8\u3060\u3051<\/td>\n\u306a\u3057<\/td>\n\u6700\u7d42\u56de\u7b54<\/td>\n\u6bd4\u8f03\u30fb\u691c\u8a3c\u7528\u3002\u69cb\u9020\u5316\u51fa\u529b\u306b\u306f\u539f\u5247\u3068\u3057\u3066\u4f7f\u308f\u306a\u3044<\/td>\n<\/tr>\n
    Assistant Prefill<\/td>\n\u306a\u3057<\/td>\n\u6700\u7d42\u56de\u7b54\u306e\u66f8\u304d\u59cb\u3081<\/td>\n\u65e2\u5b58\u5b9f\u88c5\u5411\u3051\u3002\u65b0\u898f\u5b9f\u88c5\u306b\u306f\u4f7f\u308f\u306a\u3044<\/td>\n<\/tr>\n
    JSON Schema\u51fa\u529b\u5f62\u5f0f<\/td>\n\u3042\u308a\uff08\u6b63\u5e38\u7d42\u4e86\u6642\uff09<\/td>\n\u6700\u7d42\u56de\u7b54<\/td>\n\u60c5\u5831\u62bd\u51fa\u3001\u5206\u985e\u3001API\u30ec\u30b9\u30dd\u30f3\u30b9\u306e\u7b2c\u4e00\u5019\u88dc<\/td>\n<\/tr>\n
    strict tool use<\/td>\n\u3042\u308a\uff08\u6b63\u5e38\u7d42\u4e86\u6642\uff09<\/td>\n\u30c4\u30fc\u30eb\u540d\u3068\u5165\u529b<\/td>\n\u95a2\u6570\u547c\u3073\u51fa\u3057\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3001\u5916\u90e8API\u9023\u643a<\/td>\n<\/tr>\n<\/table>\n
    \u3053\u3053\u3067\u3044\u3046\u300c\u4fdd\u8a3c\u3042\u308a\u300d\u306f\u3001\u6b63\u5e38\u7d42\u4e86\u3057\u305f\u51fa\u529b\u304cJSON Schema\u306b\u6e96\u62e0\u3059\u308b\u3088\u3046\u5236\u7d04\u3055\u308c\u308b\u3068\u3044\u3046\u610f\u5473\u3067\u3059\u3002\u30ea\u30af\u30a8\u30b9\u30c8\u304c\u5e38\u306b\u6709\u52b9\u306aJSON\u3067\u5b8c\u4e86\u3059\u308b\u3053\u3068\u3092100\uff05\u4fdd\u8a3c\u3059\u308b\u3082\u306e\u3067\u306f\u306a\u304f\u3001\u6700\u5927\u30c8\u30fc\u30af\u30f3\u6570\u3078\u306e\u5230\u9054\u3084\u3001\u540c\u3058\u5185\u5bb9\u3092\u7e70\u308a\u8fd4\u3059\u51fa\u529b\u30eb\u30fc\u30d7\u306a\u3069\u306b\u3088\u3063\u3066JSON\u304c\u672a\u5b8c\u4e86\u306b\u306a\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002\u51fa\u529b\u30eb\u30fc\u30d7\u306e\u5b9f\u4f8b\u3068\u3001\u30c8\u30fc\u30af\u30f3\u4e0a\u9650\u3001\u518d\u8a66\u884c\u3001\u76e3\u8996\u306b\u3088\u308b\u5bfe\u7b56\u306f\u3001\u4ee5\u4e0b\u306e\u8a18\u4e8b\u3067\u7d39\u4ecb\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n
    <\/div>
     \u751f\u6210AI\u95a2\u9023<\/div>
    \u751f\u6210AI\u306e\u51fa\u529b\u30eb\u30fc\u30d7\u3092\u3069\u3046\u9632\u3050\uff1f\u60f3\u5b9a\u5916\u306e\u30b3\u30b9\u30c8\u5897\u3092\u6291\u3048\u308b\u305f\u3081\u306e\u5bfe\u7b56<\/div><\/div><\/a>\n
    Amazon Bedrock\u306e\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8<\/a>\u3067\u3082\u3001JSON Schema\u51fa\u529b\u5f62\u5f0f\u3068strict tool use\u306f\u300c2\u3064\u306e\u88dc\u5b8c\u7684\u306a\u4ed5\u7d44\u307f\u300d\u3068\u8aac\u660e\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u540c\u3058\u30ea\u30af\u30a8\u30b9\u30c8\u3067\u4f75\u7528\u3059\u308b\u3053\u3068\u3082\u53ef\u80fd\u3067\u3059\u3002<\/p>\n
    \"JSON\u51fa\u529b\u3092\u5b89\u5b9a\u3055\u305b\u308b\u65b9\u6cd5\u3068Structured<\/p>\n
    JSON Schema\u51fa\u529b\u5f62\u5f0f\u3068strict tool use\u306f\u3001Structured Outputs\u306e\u4e2d\u3067\u7570\u306a\u308b\u5bfe\u8c61\u3092\u4fdd\u8a3c\u3059\u308b<\/div>\n
    \u4ee5\u4e0b\u3067\u306f\u3001Claude Haiku 4.5\u306b\u540c\u3058\u554f\u3044\u5408\u308f\u305b\u3092\u9001\u308a\u30014\u3064\u306e\u65b9\u6cd5\u3092\u6bd4\u8f03\u3057\u307e\u3059\u3002<\/p>\n
    1. \u30d7\u30ed\u30f3\u30d7\u30c8\u3060\u3051\u3067JSON\u3092\u6307\u5b9a\u3059\u308b<\/h3>\n
    \u6700\u521d\u306f\u3001\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u9805\u76ee\u3001\u5019\u88dc\u5024\u3001\u51fa\u529b\u5f62\u5f0f\u3092\u6587\u7ae0\u3067\u8a18\u8f09\u3059\u308b\u65b9\u6cd5\u3067\u3059\u3002Converse API\u3078\u901a\u5e38\u306e\u30e1\u30c3\u30bb\u30fc\u30b8\u3092\u9001\u308a\u3001\u30e2\u30c7\u30eb\u304c\u8fd4\u3057\u305f\u30c6\u30ad\u30b9\u30c8\u3092\u305d\u306e\u307e\u307e\u8868\u793a\u3057\u307e\u3059\u3002<\/p>\n
    prompt_only.py<\/div>\n
    \r\nimport os\r\n\r\nimport boto3\r\nfrom dotenv import load_dotenv\r\n\r\nPROMPT = """\r\n\u6b21\u306e\u554f\u3044\u5408\u308f\u305b\u3092JSON\u306b\u5909\u63db\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u8aac\u660e\u6587\u306f\u4ed8\u3051\u306a\u3044\u3067\u304f\u3060\u3055\u3044\u3002\r\n\r\n\u9805\u76ee:\r\n- category: \u969c\u5bb3\u3001\u8acb\u6c42\u3001\u5951\u7d04\u3001\u305d\u306e\u4ed6\u306e\u3044\u305a\u308c\u304b\r\n- priority: high\u3001medium\u3001low\u306e\u3044\u305a\u308c\u304b\r\n- summary: \u554f\u3044\u5408\u308f\u305b\u5185\u5bb9\u306e\u8981\u7d04\r\n\r\n\u554f\u3044\u5408\u308f\u305b:\r\n\u4eca\u671d\u304b\u3089\u7ba1\u7406\u753b\u9762\u3078\u30ed\u30b0\u30a4\u30f3\u3067\u304d\u307e\u305b\u3093\u3002\r\n\u696d\u52d9\u306b\u5f71\u97ff\u3059\u308b\u305f\u3081\u3001\u81f3\u6025\u78ba\u8a8d\u3092\u304a\u9858\u3044\u3057\u307e\u3059\u3002\r\n"""\r\n\r\n\r\ndef main() -> None:\r\n    load_dotenv()\r\n    client = boto3.client("bedrock-runtime", region_name=os.environ["AWS_REGION"])\r\n    response = client.converse(\r\n        modelId=os.environ["BEDROCK_MODEL_ID"],\r\n        messages=[{"role": "user", "content": [{"text": PROMPT}]}],\r\n        inferenceConfig={"maxTokens": 500, "temperature": 0},\r\n    )\r\n\r\n    print(response["output"]["message"]["content"][0]["text"])\r\n\r\n\r\nif __name__ == "__main__":\r\n    main()\r\n<\/code><\/pre>\n
    \u4eca\u56de\u306e\u5b9f\u884c\u3067\u306f\u9805\u76ee\u3068\u5024\u306f\u610f\u56f3\u3069\u304a\u308a\u3067\u3057\u305f\u304c\u3001\u300c\u8aac\u660e\u6587\u306f\u4ed8\u3051\u306a\u3044\u300d\u3068\u6307\u793a\u3057\u3066\u3082JSON\u304cMarkdown\u306e\u30b3\u30fc\u30c9\u30d6\u30ed\u30c3\u30af\u3067\u56f2\u307e\u308c\u307e\u3057\u305f\u3002<\/p>\n
    \"\u30d7\u30ed\u30f3\u30d7\u30c8\u3060\u3051\u3067JSON\u3092\u6307\u5b9a\u3057\u305f\u5b9f\u884c\u7d50\u679c\"<\/p>\n
    JSON\u306f\u5f97\u3089\u308c\u305f\u3082\u306e\u306e\u3001\u524d\u5f8c\u306b\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u304c\u4ed8\u3044\u3066\u3044\u308b<\/div>\n
    \u3053\u306e\u51fa\u529b\u3092json.loads()<\/span>\u3078\u76f4\u63a5\u6e21\u3059\u3068\u89e3\u6790\u306b\u5931\u6557\u3057\u307e\u3059\u3002\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u3092\u53d6\u308a\u9664\u3051\u3070\u89e3\u6790\u3067\u304d\u307e\u3059\u304c\u3001\u5225\u306e\u5165\u529b\u3084\u5b9f\u884c\u3067\u306f\u8aac\u660e\u6587\u3001\u9805\u76ee\u4e0d\u8db3\u3001\u5019\u88dc\u5916\u306e\u5024\u304c\u8fd4\u308b\u53ef\u80fd\u6027\u3082\u3042\u308a\u307e\u3059\u3002\u30d7\u30ed\u30f3\u30d7\u30c8\u306f\u5f62\u5f0f\u3092\u300c\u4f9d\u983c\u300d\u3057\u3066\u3044\u308b\u3060\u3051\u3067\u3042\u308a\u3001JSON\u306e\u69cb\u6587\u3084\u30b9\u30ad\u30fc\u30de\u3078\u306e\u6e96\u62e0\u3092\u4fdd\u8a3c\u3057\u3066\u3044\u307e\u305b\u3093\u3002
    \n\u305d\u306e\u305f\u3081\u3001\u69cb\u9020\u5316\u30c7\u30fc\u30bf\u3092\u5f8c\u7d9a\u51e6\u7406\u3078\u6e21\u3059\u7528\u9014\u3067\u306f\u3001\u3053\u306e\u65b9\u5f0f\u3092\u65b0\u898f\u63a1\u7528\u3057\u306a\u3044\u3053\u3068\u3092\u63a8\u5968\u3057\u307e\u3059\u3002\u307e\u305aJSON Schema\u51fa\u529b\u5f62\u5f0f\u306b\u5bfe\u5fdc\u3059\u308b\u30e2\u30c7\u30eb\u3092\u9078\u5b9a\u3057\u3001\u5229\u7528\u3059\u308bAPI\u3084\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8\u3082\u542b\u3081\u3066Structured Outputs\u3092\u5229\u7528\u3067\u304d\u308b\u69cb\u6210\u3078\u5909\u66f4\u3067\u304d\u306a\u3044\u304b\u3092\u691c\u8a0e\u3057\u307e\u3059\u3002<\/p>\n
    2. Assistant Prefill\u3067\u51fa\u529b\u3092\u8a98\u5c0e\u3059\u308b<\/h3>\n
    Assistant Prefill\u306f\u3001Structured Outputs\u767b\u5834\u524d\u306e\u65e2\u5b58\u5b9f\u88c5\u3092\u7406\u89e3\u3059\u308b\u305f\u3081\u306b\u7d39\u4ecb\u3059\u308b\u65b9\u5f0f\u3067\u3042\u308a\u3001\u65b0\u898f\u5b9f\u88c5\u306b\u306f\u63a8\u5968\u3057\u307e\u305b\u3093\u3002\u30e2\u30c7\u30eb\u306e\u56de\u7b54\u6b04\u3092\u9014\u4e2d\u307e\u3067\u66f8\u3044\u305f\u72b6\u614b\u306b\u3057\u3066\u3001\u305d\u306e\u7d9a\u304d\u3092\u751f\u6210\u3055\u305b\u308b\u65b9\u6cd5\u3067\u3059\u3002\u6b21\u306e\u4f8b\u3067\u306f\u3001assistant\u30e1\u30c3\u30bb\u30fc\u30b8\u3068\u3057\u3066\u3001\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u3068{<\/span>\u3092\u3042\u3089\u304b\u3058\u3081\u6307\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002\u3055\u3089\u306b\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u306e\u7d42\u4e86\u8a18\u53f7\u3092\u505c\u6b62\u6587\u5b57\u5217\u3068\u3057\u3066\u8a2d\u5b9a\u3057\u3001JSON\u306e\u5f8c\u306b\u6587\u7ae0\u304c\u7d9a\u304f\u3053\u3068\u3092\u6291\u3048\u307e\u3059\u3002<\/p>\n
    assistant_prefill.py<\/div>\n
    \r\nimport json\r\nimport os\r\n\r\nimport boto3\r\nfrom dotenv import load_dotenv\r\n\r\nPROMPT = """\r\n\u6b21\u306e\u554f\u3044\u5408\u308f\u305b\u3092\u3001category\u3001priority\u3001summary\u3092\u6301\u3064JSON\u306b\u5909\u63db\u3057\u3066\u304f\u3060\u3055\u3044\u3002\r\ncategory\u306f\u969c\u5bb3\u3001\u8acb\u6c42\u3001\u5951\u7d04\u3001\u305d\u306e\u4ed6\u306e\u3044\u305a\u308c\u304b\u3001\r\npriority\u306fhigh\u3001medium\u3001low\u306e\u3044\u305a\u308c\u304b\u306b\u3057\u3066\u304f\u3060\u3055\u3044\u3002\r\n\r\n\u554f\u3044\u5408\u308f\u305b:\r\n\u4eca\u671d\u304b\u3089\u7ba1\u7406\u753b\u9762\u3078\u30ed\u30b0\u30a4\u30f3\u3067\u304d\u307e\u305b\u3093\u3002\r\n\u696d\u52d9\u306b\u5f71\u97ff\u3059\u308b\u305f\u3081\u3001\u81f3\u6025\u78ba\u8a8d\u3092\u304a\u9858\u3044\u3057\u307e\u3059\u3002\r\n"""\r\n\r\n\r\ndef main() -> None:\r\n    load_dotenv()\r\n    client = boto3.client("bedrock-runtime", region_name=os.environ["AWS_REGION"])\r\n    body = {\r\n        "anthropic_version": "bedrock-2023-05-31",\r\n        "max_tokens": 500,\r\n        "temperature": 0,\r\n        "messages": [\r\n            {"role": "user", "content": PROMPT},\r\n            {"role": "assistant", "content": "```json\\n{"},\r\n        ],\r\n        "stop_sequences": ["```"],\r\n    }\r\n\r\n    response = client.invoke_model(\r\n        modelId=os.environ["BEDROCK_MODEL_ID"],\r\n        body=json.dumps(body),\r\n    )\r\n    result = json.loads(response["body"].read())\r\n    json_text = "{" + result["content"][0]["text"]\r\n\r\n    print("stop_reason:", result["stop_reason"], "\\n")\r\n    print(json.dumps(json.loads(json_text), ensure_ascii=False, indent=2))\r\n\r\n\r\nif __name__ == "__main__":\r\n    main()\r\n<\/code><\/pre>\n
    \u5b9f\u884c\u7d50\u679c\u3067\u306fstop_reason<\/span>\u304cstop_sequence<\/span>\u3068\u306a\u308a\u3001\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u3088\u308a\u524d\u3067\u751f\u6210\u304c\u505c\u6b62\u3057\u307e\u3057\u305f\u3002\u4e8b\u524d\u306b\u4e0e\u3048\u305f{<\/span>\u3092\u9023\u7d50\u3059\u308b\u3068\u3001JSON\u3068\u3057\u3066\u89e3\u6790\u3067\u304d\u307e\u3059\u3002<\/p>\n
    \"Assistant<\/p>\n
    \u4eca\u56de\u306e\u5165\u529b\u3067\u306f\u3001Assistant Prefill\u3068\u505c\u6b62\u6587\u5b57\u5217\u306b\u3088\u308a\u3001\u8aac\u660e\u6587\u3092\u542b\u307e\u306a\u3044JSON\u3092\u53d6\u5f97\u3067\u304d\u305f<\/div>\n
    \u3053\u306e\u7d50\u679c\u3060\u3051\u3092\u898b\u308b\u3068\u3001Assistant Prefill\u3067\u3082\u5341\u5206\u306b\u5b89\u5b9a\u3057\u3066\u3044\u308b\u3088\u3046\u306b\u898b\u3048\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u3002\u3057\u304b\u3057\u3001\u4eca\u56de\u306e\u4f8b\u306f\u3001\u9805\u76ee\u6570\u304c\u5c11\u306a\u304f\u5165\u529b\u3082\u77ed\u3044\u5358\u7d14\u306a\u30b1\u30fc\u30b9\u3092\u3001\u6307\u793a\u8ffd\u5f93\u6027\u80fd\u306e\u9ad8\u3044\u73fe\u5728\u306e\u30e2\u30c7\u30eb\u3067\u5b9f\u884c\u3057\u305f\u4e00\u4f8b\u306b\u3059\u304e\u307e\u305b\u3093\u3002\u9ad8\u6027\u80fd\u306a\u30e2\u30c7\u30eb\u3067\u306f\u3001\u30d7\u30ed\u30f3\u30d7\u30c8\u3060\u3051\u3001\u307e\u305f\u306fAssistant Prefill\u3067\u3082\u3001\u540c\u69d8\u306e\u5165\u529b\u306b\u5bfe\u3057\u3066\u610f\u56f3\u3069\u304a\u308a\u306eJSON\u304c\u5f97\u3089\u308c\u308b\u5834\u5408\u304c\u3042\u308a\u307e\u3059\u3002
    \n\u91cd\u8981\u306a\u306e\u306f\u3001\u300c\u4eca\u56deJSON\u3068\u3057\u3066\u89e3\u6790\u3067\u304d\u305f\u3053\u3068\u300d\u3068\u300c\u6b63\u5e38\u7d42\u4e86\u3057\u305f\u51fa\u529b\u306e\u30b9\u30ad\u30fc\u30de\u6e96\u62e0\u304c\u4ed5\u7d44\u307f\u3068\u3057\u3066\u4fdd\u8a3c\u3055\u308c\u308b\u3053\u3068\u300d\u306f\u5225\u3060\u3068\u3044\u3046\u70b9\u3067\u3059\u3002\u5165\u529b\u306e\u9577\u3055\u3084\u66d6\u6627\u3055\u3001\u9805\u76ee\u6570\u3001\u30cd\u30b9\u30c8\u306e\u6df1\u3055\u3001\u5019\u88dc\u5024\u3001\u30e2\u30c7\u30eb\u306e\u66f4\u65b0\u306a\u3069\u306b\u3088\u3063\u3066\u7d50\u679c\u306f\u5909\u5316\u3057\u5f97\u307e\u3059\u3002\u5358\u767a\u306e\u6210\u529f\u4f8b\u3060\u3051\u3092\u6839\u62e0\u306b\u672c\u756a\u63a1\u7528\u3092\u5224\u65ad\u305b\u305a\u3001\u5b9f\u30c7\u30fc\u30bf\u3092\u60f3\u5b9a\u3057\u305f\u6b63\u5e38\u7cfb\u3001\u5883\u754c\u5024\u3001\u7570\u5e38\u7cfb\u3067\u53cd\u5fa9\u8a55\u4fa1\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n
    Assistant Prefill\u3082JSON Schema\u306b\u3088\u308b\u4fdd\u8a3c\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002\u8a55\u4fa1\u6642\u306b\u9ad8\u3044\u6210\u529f\u7387\u304c\u5f97\u3089\u308c\u305f\u5834\u5408\u3067\u3082\u3001JSON\u306e\u69cb\u6587\u304c\u5d29\u308c\u308b\u3001\u5fc5\u9808\u9805\u76ee\u304c\u6b20\u3051\u308b\u3001\u578b\u304c\u7570\u306a\u308b\u3001\u5019\u88dc\u5916\u306e\u5024\u304c\u8fd4\u308b\u3068\u3044\u3063\u305f\u53ef\u80fd\u6027\u306b\u5099\u3048\u3001\u53d7\u4fe1\u5f8c\u306e\u30d1\u30fc\u30b9\u3001\u30d0\u30ea\u30c7\u30fc\u30b7\u30e7\u30f3\u3001\u518d\u8a66\u884c\u304c\u5fc5\u8981\u3067\u3059\u3002Structured Outputs\u3068\u306e\u9055\u3044\u306f\u3001\u3053\u306e\u5b9f\u884c\u7d50\u679c\u304c\u6210\u529f\u3057\u305f\u304b\u3069\u3046\u304b\u3067\u306f\u306a\u304f\u3001\u751f\u6210\u6642\u306b\u30b9\u30ad\u30fc\u30de\u6e96\u62e0\u3092\u5236\u7d04\u3067\u304d\u308b\u304b\u3069\u3046\u304b\u306b\u3042\u308a\u307e\u3059\u3002
    \n\u305f\u3060\u3057\u3001Assistant Prefill\u306f\u3001\u3059\u3079\u3066\u306e\u73fe\u884cClaude\u30e2\u30c7\u30eb\u3067\u5229\u7528\u3067\u304d\u308b\u308f\u3051\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002Anthropic\u306eMessages API\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8<\/a>\u306b\u3088\u308b\u3068\u3001Assistant Prefill\u306fClaude 4.6\u4ee5\u964d\u3067\u30b5\u30dd\u30fc\u30c8\u3055\u308c\u305a\u3001\u6307\u5b9a\u3059\u308b\u3068\u30a8\u30e9\u30fc\u306b\u306a\u308a\u307e\u3059\u3002
    \n\u65b0\u898f\u5b9f\u88c5\u3067\u306f\u3001\u307e\u305aStructured Outputs\u306b\u5bfe\u5fdc\u3059\u308b\u30e2\u30c7\u30eb\u3078\u5909\u66f4\u3067\u304d\u308b\u304b\u3092\u691c\u8a0e\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u30e2\u30c7\u30eb\u306e\u5909\u66f4\u304c\u96e3\u3057\u3044\u65e2\u5b58\u30b7\u30b9\u30c6\u30e0\u3067\u3082\u3001Assistant Prefill\u3092\u7d99\u7d9a\u5229\u7528\u3059\u308b\u524d\u306b\u3001API\u3084\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8\u306e\u5909\u66f4\u306b\u3088\u3063\u3066Structured Outputs\u3078\u79fb\u884c\u3067\u304d\u306a\u3044\u304b\u3092\u78ba\u8a8d\u3057\u307e\u3059\u3002\u3044\u305a\u308c\u3082\u5909\u66f4\u3067\u304d\u306a\u3044\u5834\u5408\u306b\u9650\u308a\u3001JSON\u306e\u7834\u640d\u3084\u30b9\u30ad\u30fc\u30de\u9055\u53cd\u304c\u8d77\u3053\u308a\u5f97\u308b\u3053\u3068\u3092\u53d7\u3051\u5165\u308c\u3001\u53d7\u4fe1\u5f8c\u306e\u30d1\u30fc\u30b9\u3001\u30d0\u30ea\u30c7\u30fc\u30b7\u30e7\u30f3\u3001\u518d\u8a66\u884c\u3001\u76e3\u8996\u3092\u7d44\u307f\u8fbc\u3093\u3060\u4e0a\u3067\u66ab\u5b9a\u7684\u306b\u5229\u7528\u3057\u307e\u3059\u3002<\/p>\n
    Error Message<\/div>\n
    \r\nbotocore.errorfactory.ValidationException: An error occurred (ValidationException) when calling the InvokeModel operation: This model does not support assistant message prefill. The conversation must end with a user message.\r\n<\/code><\/pre>\n
    3. JSON Schema\u51fa\u529b\u5f62\u5f0f\u3067\u6700\u7d42\u56de\u7b54\u3092\u4fdd\u8a3c\u3059\u308b<\/h3>\n
    \u3053\u3053\u304b\u3089\u304cAmazon Bedrock\u306eStructured Outputs\u3067\u3059\u3002JSON Schema\u51fa\u529b\u5f62\u5f0f\u3067\u306f\u3001\u30e2\u30c7\u30eb\u306e\u6700\u7d42\u56de\u7b54\u304c\u6307\u5b9a\u3057\u305f\u30b9\u30ad\u30fc\u30de\u306b\u6e96\u62e0\u3059\u308b\u3088\u3046\u3001\u751f\u6210\u53ef\u80fd\u306a\u30c8\u30fc\u30af\u30f3\u304c\u5236\u7d04\u3055\u308c\u307e\u3059\u3002
    \n\u4eca\u56de\u306f\u3001\u540c\u3058\u554f\u3044\u5408\u308f\u305b\u3092\u4f7f\u3063\u3066Pydantic\u30e2\u30c7\u30eb\u304b\u3089JSON Schema\u3092\u751f\u6210\u3057\u3001\u5b9f\u884c\u7d50\u679c\u3092\u78ba\u8a8d\u3057\u307e\u3059\u3002<\/p>\n
    json_schema_output.py<\/div>\n
    \r\nimport json\r\nimport os\r\nfrom typing import Literal\r\n\r\nimport boto3\r\nfrom dotenv import load_dotenv\r\nfrom pydantic import BaseModel, ConfigDict, Field\r\n\r\n\r\nclass SupportTicket(BaseModel):\r\n    model_config = ConfigDict(extra="forbid")\r\n\r\n    category: Literal["\u969c\u5bb3", "\u8acb\u6c42", "\u5951\u7d04", "\u305d\u306e\u4ed6"] = Field(\r\n        description="\u554f\u3044\u5408\u308f\u305b\u306e\u5206\u985e"\r\n    )\r\n    priority: Literal["high", "medium", "low"] = Field(description="\u5bfe\u5fdc\u512a\u5148\u5ea6")\r\n    summary: str = Field(description="\u554f\u3044\u5408\u308f\u305b\u5185\u5bb9\u306e\u8981\u7d04")\r\n\r\ndef main() -> None:\r\n    load_dotenv()\r\n    client = boto3.client("bedrock-runtime", region_name=os.environ["AWS_REGION"])\r\n    prompt = (\r\n        "\u6b21\u306e\u554f\u3044\u5408\u308f\u305b\u3092\u5206\u985e\u30fb\u8981\u7d04\u3057\u3066\u304f\u3060\u3055\u3044\u3002\\n\\n"\r\n        "\u4eca\u671d\u304b\u3089\u7ba1\u7406\u753b\u9762\u3078\u30ed\u30b0\u30a4\u30f3\u3067\u304d\u307e\u305b\u3093\u3002"\r\n        "\u696d\u52d9\u306b\u5f71\u97ff\u3059\u308b\u305f\u3081\u3001\u81f3\u6025\u78ba\u8a8d\u3092\u304a\u9858\u3044\u3057\u307e\u3059\u3002"\r\n    )\r\n\r\n    response = client.converse(\r\n        modelId=os.environ["BEDROCK_MODEL_ID"],\r\n        messages=[{"role": "user", "content": [{"text": prompt}]}],\r\n        inferenceConfig={"maxTokens": 500},\r\n        outputConfig={\r\n            "textFormat": {\r\n                "type": "json_schema",\r\n                "structure": {\r\n                    "jsonSchema": {\r\n                        "schema": json.dumps(\r\n                            SupportTicket.model_json_schema(), ensure_ascii=False\r\n                        ),\r\n                        "name": "support_ticket",\r\n                    }\r\n                },\r\n            }\r\n        },\r\n    )\r\n\r\n    if response["stopReason"] != "end_turn":\r\n        raise RuntimeError(f"\u751f\u6210\u304c\u6b63\u5e38\u7d42\u4e86\u3057\u307e\u305b\u3093\u3067\u3057\u305f: {response['stopReason']}")\r\n\r\n    output_text = response["output"]["message"]["content"][0]["text"]\r\n    ticket = SupportTicket.model_validate_json(output_text)\r\n    print(ticket.model_dump_json(indent=2))\r\n\r\n\r\nif __name__ == "__main__":\r\n    main()\r\n<\/code><\/pre>\n
    \u5b9f\u884c\u7d50\u679c\u306f\u3001\u30b3\u30fc\u30c9\u30d5\u30a7\u30f3\u30b9\u3092\u542b\u307e\u306a\u3044JSON\u3067\u3059\u3002category<\/span>\u3068priority<\/span>\u306fLiteral<\/span>\u3067\u6307\u5b9a\u3057\u305f\u5019\u88dc\u306b\u9650\u5b9a\u3055\u308c\u30013\u9805\u76ee\u304c\u305d\u308d\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n
    \"JSON<\/p>\n
    Pydantic\u30e2\u30c7\u30eb\u304b\u3089\u751f\u6210\u3057\u305fJSON Schema\u306b\u6e96\u62e0\u3059\u308b\u6700\u7d42\u56de\u7b54\u3092\u53d6\u5f97\u3067\u304d\u305f<\/div>\n
    \u5b9f\u88c5\u306e\u8981\u70b9\u306f\u3001\u6b21\u306e3\u70b9\u3067\u3059\u3002<\/p>\n