{"id":121871,"date":"2026-11-03T22:24:43","date_gmt":"2026-11-03T14:24:43","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/121871.html"},"modified":"2026-11-03T22:24:43","modified_gmt":"2026-11-03T14:24:43","slug":"%e7%bb%93%e6%9e%84%e5%8c%96%e8%be%93%e5%87%ba%e4%b8%8e-function-calling-%e5%b7%a5%e7%a8%8b%e5%8c%96%ef%bc%9a%e8%ae%a9%e6%a8%a1%e5%9e%8b%e8%be%93%e5%87%ba%e5%8f%af%e8%a2%ab%e6%9c%ba%e5%99%a8%e5%8f%af","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/121871.html","title":{"rendered":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39"},"content":{"rendered":"

\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316：\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39<\/h2>\n

\u672c\u6587\u628a\u300c\u8ba9\u6a21\u578b\u8f93\u51fa\u80fd\u88ab\u4ee3\u7801\u53ef\u9760\u6d88\u8d39\u300d\u62c6\u6210\u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed\u2014\u2014tool\/function calling（\u4e2d\u95f4\u52a8\u4f5c）\u4e0e structured outputs\/JSON mode（\u6700\u7ec8\u7b54\u6848），\u8bb2\u6e05\u4e3a\u4ec0\u4e48\u751f\u4ea7\u73af\u5883\u91cc\u4e00\u53e5\u300c\u8bf7\u8fd4\u56de JSON\u300d\u8fdc\u8fdc\u4e0d\u591f\u3002\u6838\u5fc3\u7ae0\u8282\u7ed9\u51fa 2026-09 \u4e09\u5927\u5382\u5546（OpenAI \/ Anthropic \/ Gemini）\u7684\u7cbe\u786e\u53c2\u6570\u3001\u7ea6\u675f\u89e3\u7801（\u91c7\u6837\u671f mask \u975e\u6cd5 token）\u4e0e\u4e8b\u540e\u6821\u9a8c（\u89e3\u6790\u540e reject）\u4e24\u79cd\u673a\u5236\u7684\u53d6\u820d，\u4ee5\u53ca\u4e00\u4efd\u4ece Pydantic\/Zod \u5b9a\u4e49\u5230\u4e1a\u52a1\u89c4\u5219\u4e8c\u6b21\u6821\u9a8c\u3001\u518d\u5230\u5355\u6b21\u91cd\u8bd5\u4e0e\u4eba\u5de5\u964d\u7ea7\u7684\u5b8c\u6574\u5de5\u7a0b\u5b9e\u73b0\u3002\u6700\u540e\u7528 6 \u6761\u8e29\u5751\u624b\u518c，\u544a\u8bc9\u4f60 additionalProperties \u6f0f\u5199\u3001strict \u4e0e best-effort \u6df7\u7528\u3001\u4f9d\u8d56 schema \u5374\u4e0d\u505a\u4e1a\u52a1\u6821\u9a8c\u8fd9\u4e9b\u4f1a\u8ba9\u4f60\u51cc\u6668\u88ab\u62a5\u8b66\u53eb\u9192\u7684\u5751\u3002\u8bfb\u5b8c\u4f60\u5e94\u80fd\u642d\u51fa\u4e00\u4e2a\u300c\u5373\u4f7f provider \u4fdd\u8bc1\u4e5f\u7559\u5ba2\u6237\u7aef\u6821\u9a8c\u300d\u7684\u7a33\u5065\u7ed3\u6784\u5316\u8f93\u51fa\u7ba1\u7ebf\u3002<\/p>\n

\u5173\u952e\u8bcd：structured outputs\u3001function calling\u3001JSON mode\u3001\u7ea6\u675f\u89e3\u7801\u3001Pydantic\u3001Zod\u3001OpenAI Responses API\u3001Anthropic output_config\u3001Gemini\u3001GBNF\u3001XGrammar\u3001Outlines\u3001BFCL<\/p>\n

\u4e00\u3001\u5199\u5728\u524d\u9762：\u4e3a\u4ec0\u4e48\u73b0\u5728\u5fc5\u987b\u804a\u8fd9\u4ef6\u4e8b<\/h3>\n

2026 \u5e74，\u5927\u6a21\u578b\u4e0d\u518d\u53ea\u662f\u300c\u804a\u5929\u6846\u91cc\u5410\u4e00\u6bb5\u5b57\u300d\u3002\u5b83\u8981\u586b\u5de5\u5355\u3001\u4e0b\u8ba2\u5355\u3001\u8c03\u63a5\u53e3\u3001\u5199\u62a5\u8868\u2014\u2014\u8fd9\u4e9b\u4e0b\u6e38\u7cfb\u7edf\u53ea\u8ba4\u7ed3\u6784\u5316\u7684\u3001\u7c7b\u578b\u6b63\u786e\u7684\u6570\u636e，\u4e0d\u8ba4\u300c\u770b\u8d77\u6765\u50cf JSON \u7684\u81ea\u7531\u6587\u672c\u300d\u3002\u4e00\u53e5\u8bdd：\u6a21\u578b\u8f93\u51fa\u80fd\u4e0d\u80fd\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39，\u76f4\u63a5\u51b3\u5b9a\u5b83\u80fd\u4e0d\u80fd\u8fdb\u751f\u4ea7\u3002<\/p>\n

\u4f46\u5f88\u591a\u56e2\u961f\u5728\u8fd9\u4ef6\u4e8b\u4e0a\u6454\u5f97\u6700\u60e8\u3002\u4ed6\u4eec\u4ee5\u4e3a prompt \u91cc\u5199\u4e00\u53e5\u300c\u8bf7\u4e25\u683c\u4ee5 JSON \u683c\u5f0f\u8fd4\u56de，\u5305\u542b\u5b57\u6bb5 a\u3001b\u3001c\u300d\u5c31\u4e07\u4e8b\u5927\u5409，\u4e0a\u7ebf\u540e\u53d1\u73b0：\u6a21\u578b\u5076\u5c14\u5728 JSON \u524d\u9762\u52a0\u4e00\u53e5\u300c\u597d\u7684，\u8fd9\u662f\u7ed3\u679c：\u300d；\u6570\u5b57\u88ab\u5f53\u6210\u5b57\u7b26\u4e32；\u5fc5\u586b\u5b57\u6bb5\u5076\u5c14\u6f0f\u6389；\u8f6c\u4e49\u5f15\u53f7\u5199\u9519\u5bfc\u81f4\u89e3\u6790\u5d29\u6e83；\u5b57\u6bb5\u540d\u88ab\u6084\u6084\u6362\u6210\u540c\u4e49\u8bcd\u3002\u6bcf\u4e00\u4e2a\u90fd\u662f\u4e00\u6b21\u751f\u4ea7\u4e8b\u6545\u3002<\/p>\n

\u8fd9\u7bc7\u6587\u7ae0\u7684\u76ee\u6807\u6709\u56db\u4e2a\u3002\u7b2c\u4e00，\u5398\u6e05\u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed：tool calling \u662f\u300c\u4e2d\u95f4\u52a8\u4f5c\u300d，structured output \u662f\u300c\u6700\u7ec8\u7b54\u6848\u300d，\u522b\u518d\u628a\u5b83\u4eec\u6df7\u4e3a\u4e00\u8c08\u3002\u7b2c\u4e8c，\u8bb2\u6e05\u4e3a\u4ec0\u4e48\u300c\u8bf7\u8fd4\u56de JSON\u300d\u5728\u751f\u4ea7\u4e0d\u591f，\u4ee5\u53ca\u7ea6\u675f\u89e3\u7801\u548c\u4e8b\u540e\u6821\u9a8c\u4e3a\u4ec0\u4e48\u8981\u7ed3\u5408\u3002\u7b2c\u4e09，\u7ed9\u51fa 2026-09 \u4e09\u5927\u5382\u5546\u7684\u7cbe\u786e\u53c2\u6570\u548c\u7ea6\u675f\u89e3\u7801\u673a\u5236，\u7167\u7740\u6284\u5c31\u80fd\u7528\u3002\u7b2c\u56db，\u7ed9\u4f60\u4e00\u5957\u5b8c\u6574\u5de5\u7a0b\u5b9e\u73b0\u548c 6 \u6761\u8e29\u5751\u624b\u518c，\u628a\u300c\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39\u300d\u4ece\u613f\u671b\u53d8\u6210 SOP\u3002<\/p>\n

\u4e00\u53e5\u8bdd\u7ed3\u8bba\u5148\u653e\u8fd9：\u7ed3\u6784\u5316\u8f93\u51fa\u4e0d\u662f prompt \u5de5\u7a0b，\u662f\u7cfb\u7edf\u5de5\u7a0b\u3002provider \u7ed9\u7684 strict \u4fdd\u8bc1\u8981\u63a5，\u4f46\u5ba2\u6237\u7aef\u6821\u9a8c\u4e00\u9053\u90fd\u4e0d\u80fd\u7701；\u6e29\u5ea6\u6ca1\u964d\u3001additionalProperties \u6f0f\u5199\u3001strict \u548c best-effort \u6df7\u7528，\u4efb\u4f55\u4e00\u4e2a\u90fd\u80fd\u8ba9\u51cc\u6668\u7684\u62a5\u8b66\u54cd\u8d77\u6765\u3002<\/p>\n

\u4e8c\u3001\u6280\u672f\u80cc\u666f\u4e0e\u73b0\u72b6\u68b3\u7406<\/h3>\n

2.1 \u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed<\/h4>\n

\u8fd9\u662f\u5168\u6587\u7684\u5730\u57fa，\u5148\u628a\u8fb9\u754c\u5212\u5e72\u51c0\u3002\u5f88\u591a\u4eba\u628a function calling \u548c JSON mode \u5f53\u6210\u540c\u4e00\u4ef6\u4e8b，\u7ed3\u679c\u5728\u67b6\u6784\u4e0a\u72af\u7cca\u6d82\u3002\u5b83\u4eec\u662f\u6b63\u4ea4\u7684，\u53ef\u4ee5\u5355\u72ec\u8981\u6c42，\u4e5f\u53ef\u4ee5\u7ec4\u5408\u8981\u6c42\u3002<\/p>\n

\u539f\u8bed\u4e00：tool \/ function calling（\u5de5\u5177\u8c03\u7528）\u3002\u6a21\u578b\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u8f93\u51fa\u4e00\u4e2a\u300c\u4e2d\u95f4\u52a8\u4f5c\u300d：\u8c03\u7528\u4e00\u4e2a\u5177\u540d\u5de5\u5177，\u5e76\u7ed9\u51fa\u5e26\u7c7b\u578b\u7684\u53c2\u6570\u3002\u5b83\u4e0d\u662f\u6700\u7ec8\u7b54\u6848，\u800c\u662f\u300c\u6211\u8981\u5e72\u8fd9\u4ef6\u4e8b\u300d\u3002\u5173\u952e\u6027\u8d28\u6709\u4e09：\u7b2c\u4e00，\u5b83\u662f\u4e2d\u95f4\u8f93\u51fa，\u6a21\u578b\u901a\u5e38\u8fd8\u4f1a\u7ee7\u7eed\u63a8\u7406\u3001\u8c03\u7528\u591a\u4e2a\u5de5\u5177\u3001\u6700\u540e\u624d\u7ed9\u6700\u7ec8\u56de\u590d；\u7b2c\u4e8c，\u53c2\u6570\u6709\u7c7b\u578b（\u6765\u81ea\u5de5\u5177 schema），\u4f46\u5de5\u5177 schema \u4e0d\u7b49\u4e8e\u6700\u7ec8\u54cd\u5e94 schema；\u7b2c\u4e09，\u53ef\u4ee5\u5e76\u884c\u2014\u2014\u4e00\u4e2a\u56de\u5408\u91cc\u6a21\u578b\u53ef\u80fd\u540c\u65f6\u51b3\u5b9a\u8c03\u7528\u591a\u4e2a\u72ec\u7acb\u5de5\u5177\u3002<\/p>\n

\u539f\u8bed\u4e8c：structured outputs \/ JSON mode（\u7ed3\u6784\u5316\u8f93\u51fa）\u3002\u6a21\u578b\u7684\u6700\u7ec8\u7b54\u6848\u88ab\u7ea6\u675f\u6210\u67d0\u4e2a JSON schema，\u4e14\u4e0d\u4ea7\u751f\u4efb\u4f55\u5de5\u5177\u8c03\u7528\u3002\u5b83\u56de\u7b54\u7684\u662f\u300c\u8fd9\u4ef6\u4e8b\u7684\u7ed3\u8bba\u662f\u4ec0\u4e48\u300d，\u800c\u4e0d\u662f\u300c\u6211\u8981\u8c03\u4ec0\u4e48\u5de5\u5177\u300d\u3002\u5b83\u7684\u4fdd\u8bc1\u901a\u5e38\u6bd4 tool calling \u66f4\u5f3a\u2014\u2014\u6bd4\u5982 OpenAI \u7684 strict \u6a21\u5f0f\u80fd\u4fdd\u8bc1\u8f93\u51fa\u5fc5\u5b9a\u7b26\u5408 schema，\u4e0d\u4f1a\u51fa\u73b0\u5b57\u6bb5\u7f3a\u5931\u6216\u7c7b\u578b\u9519\u8bef（\u5728\u4e0d\u8fdd\u53cd\u7ea6\u675f\u9650\u5236\u7684\u524d\u63d0\u4e0b）\u3002<\/p>\n

\u4e24\u8005\u600e\u4e48\u7ec4\u5408？\u4e00\u4e2a\u5178\u578b Agent \u6d41\u7a0b：\u5148\u7528 tool calling \u51b3\u5b9a\u300c\u8c03\u641c\u7d22\u3001\u8c03\u6570\u636e\u5e93\u300d，\u6700\u540e\u7528 structured outputs \u628a\u7efc\u5408\u7ed3\u8bba\u7ea6\u675f\u6210\u56fa\u5b9a schema \u8fd4\u56de\u7ed9\u4e0a\u6e38\u7cfb\u7edf\u3002\u4e5f\u53ef\u4ee5\u53ea\u8981\u6c42\u5176\u4e00：\u7eaf RAG \u95ee\u7b54\u53ea\u60f3\u8981\u7ed3\u6784\u5316\u7b54\u6848，\u5c31\u53ea\u5f00 JSON mode；\u7eaf\u5de5\u5177\u7f16\u6392\u53ea\u60f3\u8981\u52a8\u4f5c，\u5c31\u53ea\u5f00 tool calling\u3002<\/p>\n

2.2 \u4e3a\u4ec0\u4e48 BFCL \u8fd9\u7c7b\u57fa\u51c6\u91cd\u8981<\/h4>\n

tool calling \u7684\u53ef\u9760\u6027\u4e0d\u662f\u62cd\u8111\u888b\u7684，\u6709\u516c\u5f00\u57fa\u51c6\u5728\u91cf\u3002BFCL（Berkeley Function Calling Leaderboard）\u662f 2026 \u5e74\u6700\u88ab\u5e7f\u6cdb\u5f15\u7528\u7684\u5de5\u5177\u8c03\u7528\u8bc4\u6d4b，\u5b83\u628a\u300c\u6a21\u578b\u80fd\u5426\u6b63\u786e\u9009\u5de5\u5177\u3001\u586b\u5bf9\u53c2\u6570\u3001\u5904\u7406\u5e76\u884c\u4e0e\u5d4c\u5957\u300d\u62c6\u6210\u53ef\u91cf\u5316\u6307\u6807\u3002\u4e00\u4e2a\u53cd\u76f4\u89c9\u7684\u4e8b\u5b9e\u662f：\u5f88\u591a\u5728\u901a\u7528 benchmark \u4e0a\u5206\u6570\u5f88\u9ad8\u7684\u6a21\u578b，\u5728 BFCL \u4e0a\u4f1a\u7ffb\u8f66\u2014\u2014\u56e0\u4e3a\u5b83\u4eec\u64c5\u957f\u804a\u5929，\u4e0d\u64c5\u957f\u300c\u4e25\u683c\u6309 schema \u586b\u53c2\u6570\u300d\u3002\u6240\u4ee5\u9009\u578b\u65f6\u4e0d\u8981\u53ea\u770b MMLU，\u8981\u770b BFCL \u8fd9\u7c7b\u300c\u7ed3\u6784\u5316\u6d88\u8d39\u300d\u76f8\u5173\u7684\u5206\u6570\u3002<\/p>\n

2.3 2026 \u5e74\u7684\u5de5\u7a0b\u73b0\u72b6<\/h4>\n

\u51e0\u6761\u5df2\u7ecf\u5f62\u6210\u5171\u8bc6\u7684\u4e8b\u5b9e：<\/p>\n

\u7b2c\u4e00，\u300c\u8bf7\u8fd4\u56de JSON\u300d\u5f0f\u7684 prompt \u7ea6\u675f，\u672c\u8d28\u4e0a\u662f\u4e2a best-effort（\u5c3d\u529b\u800c\u4e3a）\u3002\u6a21\u578b\u5b66\u4f1a\u4e86\u300c\u901a\u5e38\u300d\u4f1a\u8fd4\u56de JSON，\u4f46\u6ca1\u6709\u4efb\u4f55\u673a\u5236\u4fdd\u8bc1\u5b83\u8fd9\u6b21\u4e00\u5b9a\u8fd4\u56de\u3002\u751f\u4ea7\u73af\u5883\u8981\u7684\u662f guarantee，\u4e0d\u662f usually\u3002<\/p>\n

\u7b2c\u4e8c，\u4e3b\u6d41\u5382\u5546\u90fd\u5728\u628a\u7ed3\u6784\u5316\u8f93\u51fa\u505a\u6210\u300c\u534f\u8bae\u7ea7\u4fdd\u8bc1\u300d\u3002OpenAI \u7684 strict\u3001Anthropic \u7684 output_config\u3001Gemini \u7684 response_json_schema，\u90fd\u662f\u5728\u91c7\u6837\u9636\u6bb5\u5c31\u4ecb\u5165，\u800c\u4e0d\u662f\u4e8b\u540e\u7948\u7977\u3002<\/p>\n

\u7b2c\u4e09，\u81ea\u6258\u7ba1\u6a21\u578b\u4e5f\u80fd\u62ff\u5230\u4fdd\u8bc1\u3002llama.cpp \u7684 GBNF\u3001XGrammar\u3001Outlines \u8fd9\u7c7b\u5de5\u5177，\u628a schema \u7f16\u8bd1\u6210 token \u7ea7\u7684\u7ea6\u675f，\u538b\u5230\u91c7\u6837\u9636\u6bb5\u505a mask，\u7b49\u4ef7\u4e8e\u5382\u5546\u7684 strict\u3002\u8fd9\u610f\u5473\u7740\u7ed3\u6784\u5316\u8f93\u51fa\u4e0d\u662f\u95ed\u6e90\u7279\u6743\u3002<\/p>\n

\u4e09\u3001\u6838\u5fc3\u539f\u7406\u62c6\u89e3<\/h3>\n

\u8fd9\u4e00\u7ae0\u662f\u5168\u6587\u6700\u91cd\u7684\u90e8\u5206，\u6211\u4eec\u6309\u300c\u4e3a\u4ec0\u4e48 prompt \u4e0d\u591f -> \u7ea6\u675f\u89e3\u7801 vs \u4e8b\u540e\u6821\u9a8c -> \u9010\u5382\u5546\u673a\u5236\u300d\u62c6\u5f00\u3002<\/p>\n

3.1 \u4e3a\u4ec0\u4e48\u300c\u8bf7\u8fd4\u56de JSON\u300d\u5728\u751f\u4ea7\u4e0d\u591f<\/h4>\n

prompt \u5f0f\u7ea6\u675f\u7684\u5931\u6548\u6a21\u5f0f\u6709\u4e94\u79cd，\u6bcf\u4e00\u79cd\u90fd\u5bf9\u5e94\u4e00\u7c7b\u771f\u5b9e\u751f\u4ea7\u4e8b\u6545：<\/p>\n

  • \u89e3\u91ca\u524d\u7f00：\u6a21\u578b\u5728 JSON \u524d\u52a0\u300c\u597d\u7684，\u4ee5\u4e0b\u662f\u7ed3\u679c：\u300d\u6216\u8005\u540e\u9762\u52a0\u300c\u5e0c\u671b\u5bf9\u4f60\u6709\u5e2e\u52a9\u300d\u3002\u89e3\u6790\u5668\u9047\u5230\u975e JSON \u5f00\u5934\u76f4\u63a5\u629b\u5f02\u5e38\u3002<\/li>\n
  • \u6f0f\u5fc5\u586b：\u67d0\u4e2a required \u5b57\u6bb5\u5076\u5c14\u6ca1\u586b，\u4e0b\u6e38 obj["x"] \u629b KeyError\u3002<\/li>\n
  • \u7c7b\u578b\u6f02\u79fb：\u6570\u5b57\u5199\u6210\u5b57\u7b26\u4e32（\u201c123\u201d \u800c\u4e0d\u662f 123），\u4e0b\u6e38\u505a\u6570\u503c\u8fd0\u7b97\u65f6\u9759\u9ed8\u51fa\u9519\u6216\u7c7b\u578b\u9519\u8bef\u3002<\/li>\n
  • \u8f6c\u4e49\u9519\u8bef：\u5b57\u6bb5\u5185\u5bb9\u91cc\u6709\u5f15\u53f7\u3001\u6362\u884c\u3001\u53cd\u659c\u6760，\u6a21\u578b\u6ca1\u6b63\u786e\u8f6c\u4e49，JSON \u89e3\u6790\u5931\u8d25\u3002<\/li>\n
  • \u5b57\u6bb5\u540d\u6f02\u79fb：\u6a21\u578b\u628a user_id \u5199\u6210 userId \u6216 userid，\u4e0b\u6e38\u53d6\u4e0d\u5230\u503c\u3002<\/li>\n

    \u8fd9\u4e94\u79cd\u91cc，\u7c7b\u578b\u6f02\u79fb\u548c\u5b57\u6bb5\u540d\u6f02\u79fb\u6700\u9634\u9669\u2014\u2014\u5b83\u4eec\u4e0d\u62a5\u9519，\u53ea\u662f\u8ba9\u4f60\u7684\u4e1a\u52a1\u903b\u8f91\u62ff\u5230 None \u6216\u9519\u8bef\u7c7b\u578b，\u7136\u540e\u6084\u65e0\u58f0\u606f\u5730\u5199\u51fa\u9519\u8bef\u6570\u636e\u3002\u8fd9\u5c31\u662f\u4e3a\u4ec0\u4e48\u7eaf\u9760 prompt \u4e0d\u591f：\u5b83\u6ca1\u6709\u300c\u786c\u7ea6\u675f\u300d，\u53ea\u6709\u300c\u8f6f\u9ed8\u5951\u300d\u3002<\/p>\n

    3.2 \u4e24\u6761\u9632\u7ebf：\u7ea6\u675f\u89e3\u7801 vs \u4e8b\u540e\u6821\u9a8c<\/h4>\n

    \u5de5\u7a0b\u4e0a\u5bf9\u4ed8\u4e0a\u8ff0\u95ee\u9898\u6709\u4e24\u6761\u8def\u7ebf，\u5b83\u4eec\u4e0d\u662f\u4e8c\u9009\u4e00，\u800c\u8981\u53e0\u52a0\u3002<\/p>\n

    \u7ea6\u675f\u89e3\u7801（constrained decoding）：\u5728 token \u91c7\u6837\u9636\u6bb5\u5c31\u4ecb\u5165\u3002\u6bcf\u5f53\u6a21\u578b\u8981\u8f93\u51fa\u4e0b\u4e00\u4e2a token，\u89e3\u7801\u5668\u6839\u636e\u5f53\u524d\u5df2\u751f\u6210\u7684 JSON \u7ed3\u6784\u548c schema，\u7b97\u51fa\u300c\u54ea\u4e9b token \u662f\u5408\u6cd5\u7684\u300d，\u628a\u975e\u6cd5 token \u7684\u91c7\u6837\u6982\u7387 mask \u6210 0\u3002\u4e8e\u662f\u6a21\u578b\u6839\u672c\u4e0d\u53ef\u80fd\u8f93\u51fa\u4e00\u4e2a\u4e0d\u7b26\u5408 schema \u7684 token \u5e8f\u5217\u3002\u8fd9\u662f\u6700\u5f3a\u4fdd\u8bc1\u2014\u2014\u8f93\u51fa\u5728\u6784\u9020\u65f6\u5c31\u5408\u6cd5\u3002OpenAI strict\u3001Anthropic strict tool schema\u3001Gemini \u7684\u53d7\u9650\u6a21\u5f0f\u3001\u81ea\u6258\u7ba1\u7684 GBNF\/XGrammar\/Outlines \u90fd\u5c5e\u4e8e\u8fd9\u4e00\u7c7b\u3002\u4ee3\u4ef7\u662f\u89e3\u7801\u5668\u53d8\u590d\u6742\u3001\u541e\u5410\u7565\u6709\u4e0b\u964d（\u8981\u5b9e\u65f6\u7b97\u5408\u6cd5 token \u96c6\u5408）\u3002<\/p>\n

    \u4e8b\u540e\u6821\u9a8c（post-hoc validation）：\u6a21\u578b\u81ea\u7531\u751f\u6210，\u751f\u6210\u5b8c\u518d parse \u548c\u6821\u9a8c\u3002\u5982\u679c\u4e0d\u7b26\u5408（\u683c\u5f0f\u9519\u3001\u7f3a\u5b57\u6bb5\u3001\u7c7b\u578b\u9519），\u5c31 reject \u5e76\u91cd\u8bd5\u6216\u964d\u7ea7\u3002\u4ee3\u4ef7\u662f\u5076\u5c14\u8981\u91cd\u8bd5，\u4e14\u5982\u679c\u6a21\u578b\u53cd\u590d\u751f\u6210\u975e\u6cd5\u8f93\u51fa\u4f1a\u6d6a\u8d39 token\u3002\u4f46\u5b83\u80fd\u515c\u4f4f\u7ea6\u675f\u89e3\u7801\u8986\u76d6\u4e0d\u5230\u7684\u300c\u4e1a\u52a1\u8bed\u4e49\u300d\u95ee\u9898（\u6bd4\u5982 quantity \u662f\u6570\u5b57\u4f46 <= 0，schema \u5141\u8bb8\u4f46\u4e1a\u52a1\u4e0d\u5141\u8bb8）\u3002<\/p>\n

    \u6b63\u786e\u505a\u6cd5\u662f\u4e24\u8005\u7ed3\u5408：\u7ea6\u675f\u89e3\u7801\u4fdd\u8bc1\u300c\u8bed\u6cd5\u5408\u6cd5\u3001\u7c7b\u578b\u5bf9\u3001\u5b57\u6bb5\u9f50\u300d，\u4e8b\u540e\u6821\u9a8c（\u5c24\u5176\u662f\u4e1a\u52a1\u89c4\u5219\u6821\u9a8c）\u4fdd\u8bc1\u300c\u8bed\u4e49\u6b63\u786e\u300d\u3002\u6280\u672f\u7ea6\u675f\u7ba1\u683c\u5f0f，\u4e1a\u52a1\u7ea6\u675f\u7ba1\u542b\u4e49，\u5404\u6709\u5404\u7684\u8fb9\u754c\u3002<\/p>\n

    3.3 \u9010\u5382\u5546\u7ea6\u675f\u89e3\u7801\u673a\u5236（2026-09）<\/h4>\n

    \u628a\u4e09\u5927\u5382\u5546\u7684\u673a\u5236\u548c\u53c2\u6570\u6446\u5728\u4e00\u5f20\u8868\u91cc：<\/p>\n\n\u5382\u5546\u7ea6\u675f\u673a\u5236\u5173\u952e\u53c2\u6570（2026-09）\u4fdd\u8bc1\u5f3a\u5ea6<\/tr>\n\n\n\n\n\n
    OpenAI<\/td>\nstrict \u7ea6\u675f\u89e3\u7801<\/td>\nResponses API: text.format={type:json_schema, strict:true}；Chat Completions \u7528 response_format<\/td>\n\u5f3a，\u8f93\u51fa\u5fc5\u7b26\u5408 schema<\/td>\n<\/tr>\n
    Anthropic<\/td>\nstrict tool schema \/ output_config<\/td>\noutput_config={format:{type:json_schema}}，\u5df2 GA \u65e0 beta<\/td>\n\u5f3a<\/td>\n<\/tr>\n
    Gemini<\/td>\nANY \u53d7\u9650\u6a21\u5f0f<\/td>\nresponse_mime_type + response_json_schema，propertyOrdering \u63a7\u5236\u987a\u5e8f<\/td>\n\u5f3a<\/td>\n<\/tr>\n
    \u81ea\u6258\u7ba1<\/td>\nGBNF \/ XGrammar \/ Outlines<\/td>\n\u628a schema \u7f16\u8bd1\u6210 token \u7ea7\u8bed\u6cd5<\/td>\n\u5f3a（\u53d6\u51b3\u4e8e\u5b9e\u73b0）<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u6ce8\u610f：\u673a\u5236\u540d\u5b57\u5404\u5bb6\u4e0d\u540c，\u4f46\u672c\u8d28\u90fd\u662f\u300c\u91c7\u6837\u671f mask \u975e\u6cd5 token\u300d\u3002\u9009\u62e9\u54ea\u5bb6\u7684\u6838\u5fc3\u5dee\u5f02\u4e0d\u5728\u673a\u5236，\u800c\u5728\u53c2\u6570\u7ec6\u8282\u548c\u7ea6\u675f\u9650\u5236\u2014\u2014\u4e0b\u4e00\u8282\u8bb2\u3002<\/p>\n

    3.4 OpenAI \u7684 strict \u53c2\u6570\u7eaa\u5f8b<\/h4>\n

    OpenAI \u7684 strict \u6a21\u5f0f（2026-09 \u4ecd\u6709\u6548）\u5bf9 schema \u6709\u4e00\u7ec4\u786c\u6027\u8981\u6c42，\u8fdd\u53cd\u4efb\u4f55\u4e00\u6761\u90fd\u4f1a\u62a5\u9519\u6216\u4e0d\u7ed9\u4fdd\u8bc1：<\/p>\n

      \n
    • additionalProperties \u5fc5\u987b\u663e\u5f0f\u8bbe\u4e3a false\u3002\u4e0d\u5199\u7b49\u4e8e\u5141\u8bb8\u989d\u5916\u5b57\u6bb5，strict \u4e0d\u4fdd\u8bc1\u5b57\u6bb5\u5c01\u95ed\u3002<\/li>\n
    • \u6240\u6709\u5b57\u6bb5\u90fd\u5fc5\u987b\u5728 required \u91cc\u3002strict \u4e0b\u6ca1\u6709\u300c\u53ef\u9009\u5b57\u6bb5\u300d\u7684\u6982\u5ff5，\u9664\u975e\u4f60\u7528 union \u628a\u7c7b\u578b\u5199\u6210 ["x", "null"] \u6765\u8868\u8fbe\u300c\u53ef\u7a7a\u300d\u3002<\/li>\n
    • \u53ef\u9009\u5b57\u6bb5\u7684\u5199\u6cd5：\u628a\u7c7b\u578b\u58f0\u660e\u4e3a ["string", "null"] \u8fd9\u7c7b\u4e24\u5143\u7d20\u6570\u7ec4，\u800c\u4e0d\u662f\u7528 required \u7701\u7565\u3002<\/li>\n
    • \u7981\u6b62\u7684 schema \u5173\u952e\u5b57：minimum \/ maximum\u3001maxLength\u3001pattern\u3001\u9012\u5f52（$ref \u81ea\u5f15\u7528 \/ \u9012\u5f52\u7ed3\u6784）\u3002\u8fd9\u4e9b\u5728 strict \u4e0b\u4e0d\u88ab\u652f\u6301，\u5199\u4e86\u4f1a\u62a5\u9519\u3002\u6240\u4ee5\u4e25\u683c\u7684\u8303\u56f4 \/ \u6b63\u5219\u7ea6\u675f\u53ea\u80fd\u653e\u5230\u4e8b\u540e\u4e1a\u52a1\u6821\u9a8c\u91cc\u505a\u3002<\/li>\n<\/ul>\n

      \u8fd9\u5957\u7eaa\u5f8b\u5f88\u53cd\u76f4\u89c9\u7684\u5730\u65b9\u662f\u300c\u53ef\u9009\u5b57\u6bb5\u8981\u7528 [\u201cx\u201d,\u201cnull\u201d]\u300d\u3002\u5f88\u591a\u4eba\u4e60\u60ef\u628a\u53ef\u9009\u5b57\u6bb5\u4ece required \u91cc\u53bb\u6389，\u4f46\u5728 strict \u4e0b\u8fd9\u4f1a\u7834\u574f\u4fdd\u8bc1\u3002\u8bb0\u4f4f：strict \u7684\u4e16\u754c\u91cc，\u8981\u4e48 required，\u8981\u4e48 [\u201ctype\u201d,\u201cnull\u201d]，\u6ca1\u6709\u7b2c\u4e09\u79cd\u300c\u53ef\u9009\u300d\u3002<\/p>\n

      3.5 Anthropic \u7684 output_config（\u5df2 GA）<\/h4>\n

      Anthropic \u5728 2026 \u5e74\u628a\u7ed3\u6784\u5316\u8f93\u51fa\u4ece beta \u7684 output_format \u63a8\u8fdb\u5230\u4e86 GA \u7684 output_config：<\/p>\n

      output_config = { "format": { "type": "json_schema", "schema": {…} } }<\/p>\n

      \u4e24\u4e2a\u8981\u70b9：\u7b2c\u4e00，\u4e0d\u8981\u518d\u4f7f\u7528\u65e7\u7684 output_format（beta \u63a5\u53e3），\u65b0\u4ee3\u7801\u7edf\u4e00\u8d70 output_config，GA \u610f\u5473\u7740\u7a33\u5b9a\u6027\u627f\u8bfa\u66f4\u5f3a\u3001\u4e0d\u4f1a\u56e0\u4e3a beta \u8c03\u6574\u800c\u53d8\u52a8\u3002\u7b2c\u4e8c，Anthropic \u7684 strict tool schema \u540c\u6837\u5728\u91c7\u6837\u671f\u505a\u7ea6\u675f，\u4fdd\u8bc1\u5de5\u5177\u53c2\u6570\u7b26\u5408 schema；\u800c output_config \u662f\u628a\u300c\u6700\u7ec8\u54cd\u5e94\u300d\u7ea6\u675f\u6210 schema，\u4e8c\u8005\u673a\u5236\u4e00\u81f4\u3001\u4f5c\u7528\u5bf9\u8c61\u4e0d\u540c（\u4e00\u4e2a\u662f\u5de5\u5177\u53c2\u6570，\u4e00\u4e2a\u662f\u6700\u7ec8\u7b54\u6848）\u3002<\/p>\n

      3.6 Gemini \u7684 response_json_schema \u4e0e propertyOrdering<\/h4>\n

      Gemini \u7684\u5199\u6cd5\u662f\u5728\u8bf7\u6c42\u91cc\u540c\u65f6\u7ed9 response_mime_type \u548c response_json_schema：<\/p>\n

      response_mime_type = "application\/json"
      \nresponse_json_schema = { "type": "object", "properties": {…}, "propertyOrdering": ["a","b","c"] }<\/p>\n

      propertyOrdering \u662f Gemini \u7279\u6709\u7684\u4e00\u4e2a\u7ec6\u8282：JSON \u5bf9\u8c61\u672c\u8eab\u662f\u65e0\u5e8f\u7684，\u4f46\u6709\u4e9b\u4e0b\u6e38\u7cfb\u7edf（\u7279\u522b\u662f\u5f3a\u987a\u5e8f\u4f9d\u8d56\u7684\u89e3\u6790\u6216\u4eba\u773c\u6bd4\u5bf9）\u5e0c\u671b\u5b57\u6bb5\u6309\u56fa\u5b9a\u987a\u5e8f\u6392\u5217\u3002propertyOrdering \u663e\u5f0f\u6307\u5b9a\u5b57\u6bb5\u8f93\u51fa\u987a\u5e8f，\u907f\u514d\u6bcf\u6b21\u987a\u5e8f\u6f02\u79fb\u5e26\u6765\u7684 diff \u566a\u58f0\u3002\u5982\u679c\u4f60 downstream \u4e0d\u5728\u4e4e\u987a\u5e8f，\u53ef\u4ee5\u4e0d\u7528；\u4f46\u5982\u679c\u4f60\u7684\u7ba1\u7ebf\u5bf9\u5b57\u6bb5\u987a\u5e8f\u654f\u611f（\u6bd4\u5982\u505a\u7b7e\u540d\u3001\u505a\u4e8c\u8fdb\u5236\u6bd4\u5bf9），\u8fd9\u4e2a\u53c2\u6570\u5f88\u5173\u952e\u3002<\/p>\n

      3.7 \u81ea\u6258\u7ba1\u7684\u7ea6\u675f\u89e3\u7801：GBNF \/ XGrammar \/ Outlines<\/h4>\n

      \u5982\u679c\u4f60\u8dd1\u7684\u662f\u5f00\u6e90\u6a21\u578b（llama.cpp\u3001vLLM \u81ea\u6258\u7ba1\u3001TGI \u7b49），\u4e5f\u80fd\u62ff\u5230\u7b49\u4ef7\u4fdd\u8bc1：<\/p>\n

        \n
      • GBNF：llama.cpp \u7684\u8bed\u6cd5\u683c\u5f0f，\u624b\u5199\u6216\u81ea\u52a8\u4ece schema \u751f\u6210\u3002\u5b83\u628a JSON schema \u7f16\u8bd1\u6210\u4e00\u4e2a\u4e0a\u4e0b\u6587\u65e0\u5173\u6587\u6cd5，\u89e3\u7801\u65f6\u53ea\u5141\u8bb8\u8d70\u5408\u6cd5\u7684\u4ea7\u751f\u5f0f\u3002<\/li>\n
      • XGrammar：\u5b57\u8282\u8df3\u52a8\u5f00\u6e90\u7684\u9ad8\u6027\u80fd\u7ea6\u675f\u89e3\u7801\u5f15\u64ce，\u628a\u8bed\u6cd5\u7f16\u8bd1\u6210\u9ad8\u6548\u7684\u4f4d\u63a9\u7801，\u541e\u5410\u635f\u5931\u5f88\u5c0f，\u88ab\u591a\u4e2a\u63a8\u7406\u6846\u67b6\u96c6\u6210\u3002<\/li>\n
      • Outlines：Python \u4fa7\u628a Pydantic \/ JSON schema \u8f6c\u6210\u7ea6\u675f，\u914d\u5408 transformers \u7cfb\u6a21\u578b\u4f7f\u7528，\u9002\u5408\u7814\u7a76\u548c\u4e2d\u5c0f\u89c4\u6a21\u90e8\u7f72\u3002<\/li>\n<\/ul>\n

        \u8fd9\u7c7b\u65b9\u6848\u7684\u4ef7\u503c\u5728\u4e8e：\u4f60\u4e0d\u9700\u8981\u4f9d\u8d56\u95ed\u6e90\u5382\u5546\u7684 strict \u5b9e\u73b0，\u4e5f\u80fd\u8ba9\u672c\u5730\u6a21\u578b\u8f93\u51fa\u300c\u8bed\u6cd5\u5408\u6cd5\u3001\u7c7b\u578b\u5bf9\u3001\u5b57\u6bb5\u9f50\u300d\u7684 JSON\u3002\u4f46\u540c\u6837\u7684，\u4e1a\u52a1\u8bed\u4e49\u6821\u9a8c\u4ecd\u8981\u81ea\u5df1\u505a\u3002<\/p>\n

        \u56db\u3001\u73af\u5883\u4e0e\u4f9d\u8d56\u51c6\u5907<\/h3>\n

        4.1 \u4f9d\u8d56\u6e05\u5355<\/h4>\n

        \u6309\u672c\u6587\u7684\u5de5\u7a0b\u5b9e\u73b0（Python \u4e3a\u4e3b，\u9644 Node\/Zod \u8bf4\u660e），\u7248\u672c\u7ec4\u5408（2026-10 \u9a8c\u8bc1）：<\/p>\n

        # Python \u4fa7：schema \u5b9a\u4e49 + \u4e09\u5927\u5382\u5546 SDK + \u6821\u9a8c<\/span>
        \npip install<\/span> -U<\/span> "pydantic>=2.7"<\/span>
        \npip install<\/span> -U<\/span> "openai>=1.50"<\/span> # Responses API \/ Chat Completions<\/span>
        \npip install<\/span> -U<\/span> "anthropic>=0.40"<\/span> # output_config<\/span>
        \n# Gemini \u901a\u8fc7 google-generativeai \u6216 openai \u517c\u5bb9\u7aef\u70b9<\/span>
        \npip install<\/span> -U<\/span> "google-generativeai>=0.8"<\/span><\/p>\n

        # Node \u4fa7（\u5982\u9700 Zod）：npm install zod<\/span>
        \n# \u81ea\u6258\u7ba1\u7ea6\u675f\u89e3\u7801（\u53ef\u9009）<\/span>
        \npip install<\/span> -U<\/span> "outlines"<\/span> # \u6216 xgrammar<\/span><\/p>\n

        \u4e24\u4e2a\u5751\u63d0\u524d\u907f：\u7b2c\u4e00，Pydantic \u4e00\u5b9a\u8981 2.x，1.x \u7684 schema \u5bfc\u51fa API \u4e0d\u540c，\u672c\u6587\u4ee3\u7801\u57fa\u4e8e v2；\u7b2c\u4e8c，OpenAI SDK \u7684 Responses API \u662f\u8f83\u65b0\u7684\u63a5\u53e3，\u65e7\u7248\u672c\u6ca1\u6709 text.format，\u88c5\u5b8c\u786e\u8ba4\u7248\u672c\u53f7\u3002<\/p>\n

        4.2 \u6700\u5c0f\u53ef\u7528\u9a8c\u8bc1<\/h4>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""\u7ed3\u6784\u5316\u8f93\u51fa\u73af\u5883\u81ea\u68c0：\u786e\u8ba4 pydantic \/ SDK \u53ef\u7528\u3002"""<\/span>
        \nimport<\/span> importlib
        \nimport<\/span> sys<\/p>\n

        def<\/span> have<\/span>(<\/span>mod)<\/span>:<\/span>
        \n try<\/span>:<\/span>
        \n importlib.<\/span>import_module(<\/span>mod)<\/span>
        \n return<\/span> True<\/span>
        \n except<\/span> Exception:<\/span>
        \n return<\/span> False<\/span><\/p>\n

        def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
        \n for<\/span> mod,<\/span> label in<\/span> (<\/span>(<\/span>"pydantic"<\/span>,<\/span> "Pydantic"<\/span>)<\/span>,<\/span>
        \n (<\/span>"openai"<\/span>,<\/span> "OpenAI SDK"<\/span>)<\/span>,<\/span>
        \n (<\/span>"anthropic"<\/span>,<\/span> "Anthropic SDK"<\/span>)<\/span>,<\/span>
        \n (<\/span>"google.generativeai"<\/span>,<\/span> "Gemini SDK"<\/span>)<\/span>)<\/span>:<\/span>
        \n print<\/span>(<\/span>"%-22s %s"<\/span> %<\/span> (<\/span>label,<\/span> "OK"<\/span> if<\/span> have(<\/span>mod)<\/span> else<\/span> "\u672a\u5b89\u88c5"<\/span>)<\/span>)<\/span>
        \n from<\/span> pydantic import<\/span> BaseModel,<\/span> Field
        \n print<\/span>(<\/span>"Pydantic \u7248\u672c\u80fd\u529b: v2 \u5bfc\u51fa JSON schema \u53ef\u7528"<\/span>)<\/span><\/p>\n

        if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
        \n main(<\/span>)<\/span><\/p>\n

        \u4e94\u3001\u5b8c\u6574\u5b9e\u64cd<\/h3>\n

        5.1 \u7b2c\u4e00\u6b65：\u7528 Pydantic \u5b9a\u4e49 schema（\u5355\u4e00\u4e8b\u5b9e\u6e90）<\/h4>\n

        \u6240\u6709\u53ef\u9760\u6027\u90fd\u4ece\u300c\u4e00\u4efd\u6743\u5a01 schema\u300d\u5f00\u59cb\u3002\u4e0d\u8981\u5728\u524d\u7aef\u3001prompt\u3001\u6821\u9a8c\u91cc\u5404\u5199\u4e00\u904d\u5b57\u6bb5，\u7528 Pydantic \u5f53\u5355\u4e00\u4e8b\u5b9e\u6e90，\u5bfc\u51fa JSON schema \u7ed9\u6a21\u578b，\u89e3\u6790\u65f6\u4e5f\u7528\u5b83\u3002<\/p>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""\u8ba2\u5355\u7ed3\u6784\u5316\u8f93\u51fa：Pydantic \u5355\u4e00\u4e8b\u5b9e\u6e90 + \u4e1a\u52a1\u89c4\u5219\u6821\u9a8c\u3002"""<\/span>
        \nfrom<\/span> enum import<\/span> Enum
        \nfrom<\/span> typing import<\/span> Optional
        \nfrom<\/span> pydantic import<\/span> BaseModel,<\/span> Field,<\/span> field_validator<\/p>\n

        class<\/span> Priority<\/span>(<\/span>str<\/span>,<\/span> Enum)<\/span>:<\/span>
        \n LOW =<\/span> "low"<\/span>
        \n NORMAL =<\/span> "normal"<\/span>
        \n HIGH =<\/span> "high"<\/span><\/p>\n

        class<\/span> OrderItem<\/span>(<\/span>BaseModel)<\/span>:<\/span>
        \n sku:<\/span> str<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> description=<\/span>"\u5546\u54c1\u7f16\u7801，\u5f62\u5982 SKU-XXXX"<\/span>)<\/span>
        \n quantity:<\/span> int<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> ge=<\/span>1<\/span>,<\/span> description=<\/span>"\u8d2d\u4e70\u6570\u91cf，\u5fc5\u987b\u4e3a\u6b63\u6574\u6570"<\/span>)<\/span>
        \n price:<\/span> float<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> description=<\/span>"\u5355\u4ef7，\u7eaf\u6570\u5b57，\u4e0d\u542b\u8d27\u5e01\u7b26\u53f7"<\/span>)<\/span><\/p>\n

        class<\/span> Order<\/span>(<\/span>BaseModel)<\/span>:<\/span>
        \n order_id:<\/span> str<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> description=<\/span>"\u8ba2\u5355\u53f7，\u5f62\u5982 ORD-YYYYMMDD-XXXX"<\/span>)<\/span>
        \n items:<\/span> list<\/span>[<\/span>OrderItem]<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> min_length=<\/span>1<\/span>)<\/span>
        \n priority:<\/span> Priority
        \n requires_review:<\/span> bool<\/span> =<\/span> Field(<\/span>.<\/span>.<\/span>.<\/span>,<\/span> description=<\/span>"\u662f\u5426\u9700\u4eba\u5de5\u590d\u6838"<\/span>)<\/span>
        \n note:<\/span> Optional[<\/span>str<\/span>]<\/span> =<\/span> None<\/span> # strict \u4e0b\u9700\u7528 ["string","null"] \u8868\u8fbe<\/span><\/p>\n

        @field_validator<\/span>(<\/span>"order_id"<\/span>)<\/span>
        \n @classmethod<\/span>
        \n def<\/span> check_order_id<\/span>(<\/span>cls,<\/span> v)<\/span>:<\/span>
        \n if<\/span> not<\/span> v.<\/span>startswith(<\/span>"ORD-"<\/span>)<\/span>:<\/span>
        \n raise<\/span> ValueError(<\/span>"order_id \u5fc5\u987b\u4ee5 ORD- \u5f00\u5934"<\/span>)<\/span>
        \n return<\/span> v<\/p>\n

        \u6ce8\u610f price \u662f float \u4e0d\u662f str\u2014\u2014\u8fd9\u6b63\u662f\u300c\u663e\u5f0f\u7c7b\u578b\u300d\u7eaa\u5f8b\u7684\u4f53\u73b0：\u8d27\u5e01\u91d1\u989d\u662f\u6570\u5b57，\u4e0d\u8981\u5728 schema \u91cc\u5199\u6210\u5e26\u8d27\u5e01\u7b26\u53f7\u7684\u5b57\u7b26\u4e32，\u5426\u5219\u4e0b\u6e38\u505a\u91d1\u989d\u8fd0\u7b97\u65f6\u4f1a\u7c7b\u578b\u5d29\u6e83\u3002<\/p>\n

        5.2 \u7b2c\u4e8c\u6b65：\u5411 OpenAI \u8bf7\u6c42 strict \u7ed3\u6784\u5316\u8f93\u51fa<\/h4>\n

        \u628a Pydantic \u5bfc\u51fa\u7684 schema \u5582\u7ed9 OpenAI \u7684 Responses API，\u5e76\u9075\u5b88 strict \u7eaa\u5f8b（additionalProperties:false\u3001\u5168\u90e8 required\u3001\u53ef\u9009\u5b57\u6bb5\u7528 [\u201cstring\u201d,\u201cnull\u201d]）：<\/p>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""OpenAI Responses API strict \u7ed3\u6784\u5316\u8f93\u51fa\u8bf7\u6c42\u9aa8\u67b6\u3002"""<\/span>
        \nimport<\/span> json
        \nfrom<\/span> openai import<\/span> OpenAI
        \nfrom<\/span> order_schema import<\/span> Order # \u89c1 5.1<\/span><\/p>\n

        def<\/span> build_strict_schema<\/span>(<\/span>model:<\/span> type<\/span>)<\/span> –<\/span>><\/span> dict<\/span>:<\/span>
        \n """\u628a Pydantic schema \u6539\u6210 strict \u53cb\u597d：\u8865 additionalProperties:false\u3002"""<\/span>
        \n schema =<\/span> model.<\/span>model_json_schema(<\/span>)<\/span>
        \n schema[<\/span>"additionalProperties"<\/span>]<\/span> =<\/span> False<\/span>
        \n # strict \u8981\u6c42\u6240\u6709\u5c5e\u6027\u8fdb required<\/span>
        \n schema[<\/span>"required"<\/span>]<\/span> =<\/span> list<\/span>(<\/span>schema.<\/span>get(<\/span>"properties"<\/span>,<\/span> {<\/span>}<\/span>)<\/span>.<\/span>keys(<\/span>)<\/span>)<\/span>
        \n return<\/span> schema<\/p>\n

        def<\/span> ask_openai<\/span>(<\/span>client:<\/span> OpenAI,<\/span> user_text:<\/span> str<\/span>)<\/span> –<\/span>><\/span> dict<\/span>:<\/span>
        \n resp =<\/span> client.<\/span>responses.<\/span>create(<\/span>
        \n model=<\/span>"gpt-4o-mini"<\/span>,<\/span>
        \n input<\/span>=<\/span>user_text,<\/span>
        \n text=<\/span>{<\/span>
        \n "format"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "json_schema"<\/span>,<\/span>
        \n "name"<\/span>:<\/span> "order"<\/span>,<\/span>
        \n "strict"<\/span>:<\/span> True<\/span>,<\/span>
        \n "schema"<\/span>:<\/span> build_strict_schema(<\/span>Order)<\/span>,<\/span>
        \n }<\/span>
        \n }<\/span>,<\/span>
        \n temperature=<\/span>0.0<\/span>,<\/span> # \u7ed3\u6784\u5316\u8f93\u51fa\u5fc5\u987b\u964d\u6e29，\u89c1\u7b2c\u4e03\u7ae0<\/span>
        \n )<\/span>
        \n return<\/span> json.<\/span>loads(<\/span>resp.<\/span>output_text)<\/span><\/p>\n

        if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
        \n c =<\/span> OpenAI(<\/span>)<\/span>
        \n out =<\/span> ask_openai(<\/span>c,<\/span> "\u7528\u6237\u60f3\u4e70 2 \u4ef6 SKU-A100，\u5355\u4ef7 9.9，\u52a0\u6025，\u9700\u590d\u6838"<\/span>)<\/span>
        \n print<\/span>(<\/span>json.<\/span>dumps(<\/span>out,<\/span> ensure_ascii=<\/span>False<\/span>,<\/span> indent=<\/span>2<\/span>)<\/span>)<\/span><\/p>\n

        temperature=0.0 \u4e0d\u662f\u53ef\u9009\u9879\u2014\u2014\u7b2c\u4e03\u7ae0\u7b2c\u4e94\u5751\u4f1a\u8bb2，\u6e29\u5ea6\u6ca1\u964d\u5076\u53d1\u683c\u5f0f\u9519\u3002<\/p>\n

        5.3 \u7b2c\u4e09\u6b65：\u5411 Anthropic \u8bf7\u6c42 output_config<\/h4>\n

        Anthropic \u8d70 GA \u7684 output_config，\u4e0d\u518d\u7528 beta \u7684 output_format：<\/p>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""Anthropic output_config \u7ed3\u6784\u5316\u8f93\u51fa\u8bf7\u6c42\u9aa8\u67b6\u3002"""<\/span>
        \nimport<\/span> json
        \nfrom<\/span> anthropic import<\/span> Anthropic
        \nfrom<\/span> order_schema import<\/span> Order<\/p>\n

        def<\/span> ask_anthropic<\/span>(<\/span>client:<\/span> Anthropic,<\/span> user_text:<\/span> str<\/span>)<\/span> –<\/span>><\/span> dict<\/span>:<\/span>
        \n resp =<\/span> client.<\/span>messages.<\/span>create(<\/span>
        \n model=<\/span>"claude-sonnet-4-5"<\/span>,<\/span>
        \n max_tokens=<\/span>1024<\/span>,<\/span>
        \n temperature=<\/span>0.0<\/span>,<\/span>
        \n messages=<\/span>[<\/span>{<\/span>"role"<\/span>:<\/span> "user"<\/span>,<\/span> "content"<\/span>:<\/span> user_text}<\/span>]<\/span>,<\/span>
        \n output_config=<\/span>{<\/span>
        \n "format"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "json_schema"<\/span>,<\/span>
        \n "schema"<\/span>:<\/span> Order.<\/span>model_json_schema(<\/span>)<\/span>,<\/span>
        \n }<\/span>
        \n }<\/span>,<\/span>
        \n )<\/span>
        \n # output_config \u4fdd\u8bc1 text \u5373\u5408\u6cd5 JSON<\/span>
        \n return<\/span> json.<\/span>loads(<\/span>resp.<\/span>content[<\/span>0<\/span>]<\/span>.<\/span>text)<\/span><\/p>\n

        if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
        \n c =<\/span> Anthropic(<\/span>)<\/span>
        \n out =<\/span> ask_anthropic(<\/span>c,<\/span> "\u4e0b\u5355：3 \u4ef6 SKU-B200，\u5355\u4ef7 19.9，\u666e\u901a\u4f18\u5148\u7ea7"<\/span>)<\/span>
        \n print<\/span>(<\/span>json.<\/span>dumps(<\/span>out,<\/span> ensure_ascii=<\/span>False<\/span>,<\/span> indent=<\/span>2<\/span>)<\/span>)<\/span><\/p>\n

        5.4 \u7b2c\u56db\u6b65：Gemini \u7684 response_json_schema \u8bf7\u6c42\u4f53<\/h4>\n

        Gemini \u7528 response_mime_type + response_json_schema，\u5e76\u7528 propertyOrdering \u56fa\u5b9a\u5b57\u6bb5\u987a\u5e8f：<\/p>\n

        {<\/span>
        \n "contents"<\/span>:<\/span> [<\/span>
        \n {<\/span> "role"<\/span>:<\/span> "user"<\/span>,<\/span> "parts"<\/span>:<\/span> [<\/span>{<\/span> "text"<\/span>:<\/span> "\u4e0b\u5355：1 \u4ef6 SKU-C300，\u5355\u4ef7 5.0，\u9ad8\u4f18\u5148\u7ea7，\u9700\u590d\u6838"<\/span> }<\/span>]<\/span> }<\/span>
        \n ]<\/span>,<\/span>
        \n "generationConfig"<\/span>:<\/span> {<\/span>
        \n "responseMimeType"<\/span>:<\/span> "application\/json"<\/span>,<\/span>
        \n "responseSchema"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "OBJECT"<\/span>,<\/span>
        \n "properties"<\/span>:<\/span> {<\/span>
        \n "order_id"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "STRING"<\/span> }<\/span>,<\/span>
        \n "items"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "ARRAY"<\/span>,<\/span>
        \n "items"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "OBJECT"<\/span>,<\/span>
        \n "properties"<\/span>:<\/span> {<\/span>
        \n "sku"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "STRING"<\/span> }<\/span>,<\/span>
        \n "quantity"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "INTEGER"<\/span> }<\/span>,<\/span>
        \n "price"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "NUMBER"<\/span> }<\/span>
        \n }<\/span>
        \n }<\/span>
        \n }<\/span>,<\/span>
        \n "priority"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "STRING"<\/span>,<\/span> "enum"<\/span>:<\/span> [<\/span>"low"<\/span>,<\/span> "normal"<\/span>,<\/span> "high"<\/span>]<\/span> }<\/span>,<\/span>
        \n "requires_review"<\/span>:<\/span> {<\/span> "type"<\/span>:<\/span> "BOOLEAN"<\/span> }<\/span>
        \n }<\/span>,<\/span>
        \n "propertyOrdering"<\/span>:<\/span> [<\/span>"order_id"<\/span>,<\/span> "items"<\/span>,<\/span> "priority"<\/span>,<\/span> "requires_review"<\/span>]<\/span>
        \n }<\/span>
        \n }<\/span>
        \n}<\/span><\/p>\n

        5.5 \u7b2c\u4e94\u6b65：\u89e3\u6790 + \u4e1a\u52a1\u89c4\u5219\u4e8c\u6b21\u6821\u9a8c + \u5355\u6b21\u91cd\u8bd5 + \u4eba\u5de5\u964d\u7ea7<\/h4>\n

        \u8fd9\u662f\u5168\u6587\u6700\u8be5\u88ab\u6284\u8d70\u7684\u5de5\u7a0b\u9aa8\u67b6\u3002\u7ea6\u675f\u89e3\u7801\u4fdd\u8bc1\u300c\u683c\u5f0f\u5bf9\u300d，\u4f46\u300c\u4e1a\u52a1\u5bf9\u300d\u8981\u9760\u8fd9\u91cc：<\/p>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""\u7ed3\u6784\u5316\u8f93\u51fa\u5b8c\u6574\u7ba1\u7ebf：\u89e3\u6790 -> \u4e1a\u52a1\u6821\u9a8c -> \u5355\u6b21\u7d27\u5316\u91cd\u8bd5 -> \u4eba\u5de5\u964d\u7ea7\u3002"""<\/span>
        \nimport<\/span> json
        \nimport<\/span> sys
        \nfrom<\/span> order_schema import<\/span> Order,<\/span> Priority<\/p>\n

        def<\/span> parse_and_validate<\/span>(<\/span>raw:<\/span> str<\/span>,<\/span> user_text:<\/span> str<\/span>)<\/span> –<\/span>><\/span> Order:<\/span>
        \n """\u89e3\u6790 JSON \u5e76\u505a Pydantic + \u4e1a\u52a1\u89c4\u5219\u6821\u9a8c，\u5931\u8d25\u629b\u5f02\u5e38\u3002"""<\/span>
        \n data =<\/span> json.<\/span>loads(<\/span>raw)<\/span> # \u683c\u5f0f\u9519\u5728\u6b64\u629b\u51fa<\/span>
        \n order =<\/span> Order(<\/span>**<\/span>data)<\/span> # \u7c7b\u578b\/\u5fc5\u586b\/\u5b57\u6bb5\u540d\u5728\u6b64\u4fdd\u8bc1<\/span>
        \n # \u4e1a\u52a1\u89c4\u5219\u4e8c\u6b21\u6821\u9a8c（schema \u7ba1\u4e0d\u4e86\u8bed\u4e49）<\/span>
        \n total =<\/span> sum<\/span>(<\/span>i.<\/span>price *<\/span> i.<\/span>quantity for<\/span> i in<\/span> order.<\/span>items)<\/span>
        \n if<\/span> total <=<\/span> 0<\/span>:<\/span>
        \n raise<\/span> ValueError(<\/span>"\u8ba2\u5355\u603b\u91d1\u989d\u5fc5\u987b\u4e3a\u6b63"<\/span>)<\/span>
        \n if<\/span> order.<\/span>requires_review and<\/span> order.<\/span>priority ==<\/span> Priority.<\/span>LOW:<\/span>
        \n raise<\/span> ValueError(<\/span>"\u9700\u590d\u6838\u7684\u8ba2\u5355\u4e0d\u5e94\u662f low \u4f18\u5148\u7ea7"<\/span>)<\/span>
        \n return<\/span> order<\/p>\n

        def<\/span> pipeline<\/span>(<\/span>ask_fn,<\/span> user_text:<\/span> str<\/span>,<\/span> max_retry:<\/span> int<\/span> =<\/span> 1<\/span>)<\/span> –<\/span>><\/span> Order:<\/span>
        \n last_err =<\/span> None<\/span>
        \n for<\/span> attempt in<\/span> range<\/span>(<\/span>max_retry +<\/span> 1<\/span>)<\/span>:<\/span>
        \n try<\/span>:<\/span>
        \n raw =<\/span> ask_fn(<\/span>user_text,<\/span> feedback=<\/span>str<\/span>(<\/span>last_err)<\/span> if<\/span> last_err else<\/span> None<\/span>)<\/span>
        \n return<\/span> parse_and_validate(<\/span>raw,<\/span> user_text)<\/span>
        \n except<\/span> Exception as<\/span> exc:<\/span> # noqa: BLE001<\/span>
        \n last_err =<\/span> exc
        \n print<\/span>(<\/span>"[retry %d] \u6821\u9a8c\u5931\u8d25: %s"<\/span> %<\/span> (<\/span>attempt,<\/span> exc)<\/span>)<\/span>
        \n # \u515c\u5e95：\u964d\u7ea7\u4eba\u5de5\u961f\u5217，\u7edd\u4e0d\u76f4\u63a5\u628a\u810f\u6570\u636e\u5199\u5e93<\/span>
        \n print<\/span>(<\/span>"\u964d\u7ea7\u4eba\u5de5\u961f\u5217: %s"<\/span> %<\/span> user_text)<\/span>
        \n raise<\/span> RuntimeError(<\/span>"structured output \u591a\u6b21\u5931\u8d25，\u5df2\u8f6c\u4eba\u5de5"<\/span>)<\/span><\/p>\n

        if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
        \n # \u793a\u4f8b ask_fn：\u771f\u5b9e\u73af\u5883\u91cc\u66ff\u6362\u4e3a 5.2\/5.3\/5.4 \u7684\u8c03\u7528<\/span>
        \n def<\/span> fake_ask<\/span>(<\/span>text,<\/span> feedback=<\/span>None<\/span>)<\/span>:<\/span>
        \n return<\/span> (<\/span>'{"order_id":"ORD-20261003-0001",'<\/span>
        \n '"items":[{"sku":"SKU-A100","quantity":2,"price":9.9}],'<\/span>
        \n '"priority":"high","requires_review":true}'<\/span>)<\/span>
        \n order =<\/span> pipeline(<\/span>fake_ask,<\/span> "\u7528\u6237\u4e0b\u5355"<\/span>)<\/span>
        \n print<\/span>(<\/span>"\u89e3\u6790\u6210\u529f:"<\/span>,<\/span> order.<\/span>order_id,<\/span> "\u603b\u989d"<\/span>,<\/span>
        \n sum<\/span>(<\/span>i.<\/span>price *<\/span> i.<\/span>quantity for<\/span> i in<\/span> order.<\/span>items)<\/span>)<\/span><\/p>\n

        \u5173\u952e\u8bbe\u8ba1：provider \u7ed9\u4e86 strict \u4fdd\u8bc1，\u6211\u4ecd\u7136\u5728\u5ba2\u6237\u7aef\u7528 Pydantic \u518d\u9a8c\u4e00\u904d\u3002\u56e0\u4e3a strict \u4e0d\u8986\u76d6\u4e1a\u52a1\u8bed\u4e49（quantity>0\u3001order_id \u683c\u5f0f\u3001\u4f18\u5148\u7ea7\u4e0e\u590d\u6838\u7684\u4e00\u81f4\u6027），\u800c\u8fd9\u4e9b\u90fd\u662f\u810f\u6570\u636e\u8fdb\u5e93\u7684\u9ad8\u9891\u6765\u6e90\u3002\u5373\u4f7f\u6709\u4e00\u5929 provider \u5347\u7ea7\u7ed5\u8fc7\u4e86\u4fdd\u8bc1（\u7b2c\u4e03\u7ae0\u7b2c\u56db\u5751），\u5ba2\u6237\u7aef\u8fd9\u5c42\u4ecd\u80fd\u515c\u5e95\u3002<\/p>\n

        5.6 \u7b2c\u516d\u6b65：tool calling \u4e0e structured output \u7ec4\u5408<\/h4>\n

        \u4e00\u4e2a Agent \u65e2\u8981\u7528\u5de5\u5177\u3001\u6700\u540e\u53c8\u8981\u7ed3\u6784\u5316\u7b54\u6848\u65f6，\u7ec4\u5408\u5199\u6cd5（\u4ee5 OpenAI \u4e3a\u4f8b）：<\/p>\n

        #!\/usr\/bin\/env python3<\/span>
        \n# -*- coding: utf-8 -*-<\/span>
        \n"""\u7ec4\u5408\u793a\u4f8b：\u5148\u7528 tool calling \u67e5\u5e93\u5b58，\u518d\u7528 structured output \u8fd4\u56de\u7ed3\u8bba\u3002"""<\/span>
        \nimport<\/span> json
        \nfrom<\/span> openai import<\/span> OpenAI<\/p>\n

        def<\/span> ask_with_tools<\/span>(<\/span>client:<\/span> OpenAI,<\/span> user_text:<\/span> str<\/span>)<\/span> –<\/span>><\/span> dict<\/span>:<\/span>
        \n # \u7b2c\u4e00\u6b65：\u5f00 tool calling，\u8ba9\u6a21\u578b\u51b3\u5b9a\u67e5\u5e93\u5b58<\/span>
        \n resp =<\/span> client.<\/span>chat.<\/span>completions.<\/span>create(<\/span>
        \n model=<\/span>"gpt-4o-mini"<\/span>,<\/span>
        \n messages=<\/span>[<\/span>{<\/span>"role"<\/span>:<\/span> "user"<\/span>,<\/span> "content"<\/span>:<\/span> user_text}<\/span>]<\/span>,<\/span>
        \n tools=<\/span>[<\/span>{<\/span>
        \n "type"<\/span>:<\/span> "function"<\/span>,<\/span>
        \n "function"<\/span>:<\/span> {<\/span>
        \n "name"<\/span>:<\/span> "check_stock"<\/span>,<\/span>
        \n "parameters"<\/span>:<\/span> {<\/span>
        \n "type"<\/span>:<\/span> "object"<\/span>,<\/span>
        \n "properties"<\/span>:<\/span> {<\/span>"sku"<\/span>:<\/span> {<\/span>"type"<\/span>:<\/span> "string"<\/span>}<\/span>}<\/span>,<\/span>
        \n "required"<\/span>:<\/span> [<\/span>"sku"<\/span>]<\/span>,<\/span>
        \n }<\/span>,<\/span>
        \n }<\/span>,<\/span>
        \n }<\/span>]<\/span>,<\/span>
        \n )<\/span>
        \n # \u7b2c\u4e8c\u6b65：\u62ff\u5230\u5de5\u5177\u7ed3\u679c\u540e，\u518d\u8981\u6c42 structured output \u8fd4\u56de\u6700\u7ec8\u7ed3\u8bba<\/span>
        \n # （\u6b64\u5904\u7701\u7565\u7b2c\u4e8c\u6b21\u8c03\u7528\u7684 json_schema \u7ea6\u675f，\u590d\u7528 5.2 \u7684 build_strict_schema）<\/span>
        \n return<\/span> resp.<\/span>choices[<\/span>0<\/span>]<\/span>.<\/span>message<\/p>\n

        if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
        \n c =<\/span> OpenAI(<\/span>)<\/span>
        \n msg =<\/span> ask_with_tools(<\/span>c,<\/span> "SKU-A100 \u6709\u8d27\u5417？\u7ed9\u6211\u4e00\u4e2a\u7ed3\u6784\u5316\u7ed3\u8bba"<\/span>)<\/span>
        \n print<\/span>(<\/span>json.<\/span>dumps(<\/span>msg.<\/span>model_dump(<\/span>)<\/span>,<\/span> ensure_ascii=<\/span>False<\/span>,<\/span> indent=<\/span>2<\/span>)<\/span>)<\/span><\/p>\n

        \u516d\u3001\u7ed3\u679c\u5bf9\u6bd4\u4e0e\u6570\u636e\u5206\u6790<\/h3>\n

        6.1 \u4e09\u79cd\u4fdd\u8bc1\u5f3a\u5ea6\u7684\u5bf9\u7167<\/h4>\n

        \u628a\u300c\u7eaf prompt\u300d\u300c\u7ea6\u675f\u89e3\u7801\u300d\u300c\u7ea6\u675f\u89e3\u7801 + \u4e1a\u52a1\u6821\u9a8c\u300d\u4e09\u79cd\u65b9\u6848\u653e\u5728\u540c\u4e00\u7ef4\u5ea6\u6bd4：<\/p>\n\n\u65b9\u6848\u683c\u5f0f\u5408\u6cd5\u7c7b\u578b\u6b63\u786e\u5b57\u6bb5\u9f50\u4e1a\u52a1\u8bed\u4e49\u6b63\u786e\u989d\u5916\u6210\u672c<\/tr>\n\n\n\n\n
        \u7eaf prompt\u300c\u8bf7\u8fd4\u56de JSON\u300d<\/td>\n\u5076\u5c14\u5931\u8d25<\/td>\n\u5076\u5c14\u6f02\u79fb<\/td>\n\u5076\u5c14\u6f0f<\/td>\n\u4e0d\u4fdd\u8bc1<\/td>\n\u51e0\u4e4e\u65e0<\/td>\n<\/tr>\n
        \u7ea6\u675f\u89e3\u7801（strict\/GBNF）<\/td>\n\u4fdd\u8bc1<\/td>\n\u4fdd\u8bc1<\/td>\n\u4fdd\u8bc1<\/td>\n\u4e0d\u4fdd\u8bc1<\/td>\n\u89e3\u7801\u7565\u6162<\/td>\n<\/tr>\n
        \u7ea6\u675f\u89e3\u7801 + \u4e1a\u52a1\u6821\u9a8c<\/td>\n\u4fdd\u8bc1<\/td>\n\u4fdd\u8bc1<\/td>\n\u4fdd\u8bc1<\/td>\n\u4fdd\u8bc1<\/td>\n\u89e3\u7801\u7565\u6162 + \u6821\u9a8c\u4ee3\u7801<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u7ed3\u8bba\u5f88\u6e05\u695a：\u53ea\u6709\u7b2c\u4e09\u884c\u80fd\u8fdb\u751f\u4ea7\u3002\u7b2c\u4e00\u884c\u662f demo \u7ea7\u522b，\u7b2c\u4e8c\u884c\u80fd\u6321\u4f4f\u8bed\u6cd5\/\u7c7b\u578b\u4e8b\u6545\u4f46\u6321\u4e0d\u4f4f\u300cquantity=-1\u300d\u8fd9\u79cd\u8bed\u4e49\u4e8b\u6545\u3002<\/p>\n

        6.2 \u7ea6\u675f\u89e3\u7801 vs \u4e8b\u540e\u6821\u9a8c\u7684\u53d6\u820d<\/h4>\n\n\u7ef4\u5ea6\u7ea6\u675f\u89e3\u7801\u4e8b\u540e\u6821\u9a8c<\/tr>\n\n\n\n\n\n\n
        \u5931\u8d25\u65f6\u673a<\/td>\n\u6784\u9020\u671f\u5c31\u5408\u6cd5<\/td>\n\u751f\u6210\u5b8c\u624d\u53d1\u73b0<\/td>\n<\/tr>\n
        \u80fd\u5426 100% \u4fdd\u8bc1\u683c\u5f0f<\/td>\n\u80fd（\u5728\u7ea6\u675f\u9650\u5236\u5185）<\/td>\n\u4e0d\u80fd，\u4f9d\u8d56 retry<\/td>\n<\/tr>\n
        \u80fd\u5426\u8986\u76d6\u4e1a\u52a1\u8bed\u4e49<\/td>\n\u4e0d\u80fd（schema \u4e0d\u7ba1\u8bed\u4e49）<\/td>\n\u80fd（\u5199\u4e1a\u52a1\u89c4\u5219）<\/td>\n<\/tr>\n
        \u541e\u5410\u5f71\u54cd<\/td>\n\u7565\u964d（\u7b97\u5408\u6cd5 token）<\/td>\n\u91cd\u8bd5\u65f6\u6d6a\u8d39 token<\/td>\n<\/tr>\n
        \u9002\u7528<\/td>\n\u6240\u6709\u7ed3\u6784\u5316\u8f93\u51fa<\/td>\n\u4f5c\u4e3a\u515c\u5e95\u53e0\u52a0<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u6240\u4ee5\u4e8c\u8005\u662f\u4e92\u8865\u4e0d\u662f\u66ff\u4ee3：\u7ea6\u675f\u89e3\u7801\u505a\u300c\u7b2c\u4e00\u9053\u8bed\u6cd5\u95f8\u300d，\u4e8b\u540e\u6821\u9a8c\u505a\u300c\u7b2c\u4e8c\u9053\u8bed\u4e49\u95f8\u300d\u3002<\/p>\n

        6.3 \u9010\u5382\u5546\u53c2\u6570\u901f\u67e5\u8868<\/h4>\n\n\u5382\u5546\u63a5\u53e3\u4f4d\u7f6estrict\/\u7ea6\u675f\u5f00\u5173\u5fc5\u5199\u7eaa\u5f8b\u7981\u6b62\u5173\u952e\u5b57<\/tr>\n\n\n\n\n
        OpenAI<\/td>\nResponses: text.format；CC: response_format<\/td>\nstrict:true<\/td>\nadditionalProperties:false\u3001\u5168 required\u3001\u53ef\u9009\u7528 [\u201ct\u201d,\u201cnull\u201d]<\/td>\nminimum\/maxLength\/pattern\/\u9012\u5f52<\/td>\n<\/tr>\n
        Anthropic<\/td>\noutput_config.format<\/td>\n\u9ed8\u8ba4\u7ea6\u675f<\/td>\n\u8d70 json_schema<\/td>\n\u540c OpenAI \u7c7b\u7ea6\u675f<\/td>\n<\/tr>\n
        Gemini<\/td>\nresponse_mime_type+response_schema<\/td>\nANY \u53d7\u9650<\/td>\n\u7528 propertyOrdering \u63a7\u987a\u5e8f<\/td>\n\u89c6\u5b9e\u73b0<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        6.4 \u4e00\u4e2a\u771f\u5b9e\u4e8b\u6545\u7684\u91cf\u7ea7<\/h4>\n

        community \u91cc\u53cd\u590d\u51fa\u73b0\u7684\u6848\u4f8b：\u67d0\u8ba2\u5355\u7cfb\u7edf\u7528\u7eaf prompt \u7ea6\u675f，\u4e0a\u7ebf\u9996\u5468\u7ea6 3% \u7684\u8bf7\u6c42\u4ea7\u51fa\u975e\u6cd5 JSON \u6216\u5b57\u6bb5\u6f02\u79fb，\u5176\u4e2d\u7ea6\u4e00\u534a\u662f\u300c\u7c7b\u578b\u6f02\u79fb\u300d（\u4ef7\u683c\u53d8\u5b57\u7b26\u4e32），\u5bfc\u81f4\u4e0b\u6e38\u5bf9\u8d26\u7cfb\u7edf\u628a\u4e00\u6279\u8ba2\u5355\u91d1\u989d\u7b97\u6210 0\u3002\u6362\u6210 strict + \u4e1a\u52a1\u6821\u9a8c\u540e，\u683c\u5f0f\u4e8b\u6545\u5f52\u96f6，\u4e1a\u52a1\u8bed\u4e49\u4e8b\u6545（quantity<=0\u3001order_id \u683c\u5f0f\u9519）\u7531\u6821\u9a8c\u5c42\u62e6\u622a，\u8f6c\u4eba\u5de5\u5904\u7406\u30023% \u542c\u8d77\u6765\u4e0d\u5927，\u4f46\u5728\u65e5\u5747\u767e\u4e07\u8ba2\u5355\u7684\u4f53\u91cf\u4e0b\u662f\u6bcf\u5929\u4e09\u4e07\u6b21\u810f\u6570\u636e\u2014\u2014\u8db3\u4ee5\u8ba9\u8d22\u52a1\u5bf9\u8d26\u5d29\u6e83\u3002<\/p>\n

        \u4e03\u3001\u8e29\u5751\u8bb0\u5f55\u4e0e\u6392\u67e5\u624b\u518c<\/h3>\n

        7.1 \u8e29\u5751 1：\u6f0f\u5199 additionalProperties:false，\u5192\u51fa\u591a\u4f59\u5b57\u6bb5<\/h4>\n

        \u73b0\u8c61：strict \u6a21\u5f0f\u4e0b，\u6a21\u578b\u5076\u5c14\u591a\u8fd4\u56de\u4e00\u4e2a\u6587\u6863\u91cc\u6ca1\u5b9a\u4e49\u7684\u5b57\u6bb5（\u6bd4\u5982\u591a\u585e\u4e00\u4e2a confidence），\u4e0b\u6e38\u7cfb\u7edf\u867d\u7136\u5927\u591a\u5ffd\u7565，\u4f46\u6709\u51e0\u4e2a\u4e25\u683c\u6821\u9a8c\u7684\u6d88\u8d39\u8005\u76f4\u63a5\u62a5\u9519\u3002<\/p>\n

        \u5b9a\u4f4d：\u770b\u53d1\u51fa\u7684 schema，\u53d1\u73b0 additionalProperties \u6ca1\u663e\u5f0f\u5199 false\u3002OpenAI strict \u8981\u6c42\u663e\u5f0f\u58f0\u660e，\u4e0d\u5199\u7b49\u4e8e\u5141\u8bb8\u989d\u5916\u5b57\u6bb5，\u4fdd\u8bc1\u88ab\u7834\u574f\u3002<\/p>\n

        \u539f\u56e0：\u5f88\u591a\u4eba\u4ee5\u4e3a strict \u9ed8\u8ba4\u5c31\u5c01\u95ed\u5b57\u6bb5，\u5b9e\u9645\u4e0a\u5fc5\u987b\u663e\u5f0f additionalProperties:false\u3002\u6f0f\u5199\u540e，\u6a21\u578b\u53ef\u4ee5\u5728 schema \u4e4b\u5916\u81ea\u7531\u52a0\u5b57\u6bb5，\u7834\u574f\u300c\u5b57\u6bb5\u5c01\u95ed\u300d\u7684\u627f\u8bfa\u3002<\/p>\n

        \u89e3\u51b3：schema \u751f\u6210\u51fd\u6570\u91cc\u5f3a\u5236\u8865 schema["additionalProperties"] = False，\u5e76\u5728\u5355\u6d4b\u91cc\u65ad\u8a00\u3002\u8fd9\u662f\u4e00\u884c\u4ee3\u7801，\u4f46\u6f0f\u4e86\u5c31\u662f\u4e00\u7c7b\u4e8b\u6545\u3002<\/p>\n

        7.2 \u8e29\u5751 2：\u5d4c\u5957\u5c42\u7ea7 \/ \u5c5e\u6027\u6570\u8d85\u9650<\/h4>\n

        \u73b0\u8c61：\u628a\u4e00\u4e2a\u5f88\u6df1\u7684\u5d4c\u5957 schema（\u6bd4\u5982 6 \u5c42\u5d4c\u5957\u3001\u5355\u5bf9\u8c61 60 \u4e2a\u5c5e\u6027）\u4ea4\u7ed9 strict，\u8fd4\u56de\u62a5\u9519\u6216\u6a21\u578b\u5410\u51fa\u622a\u65ad\/\u7578\u5f62\u7684 JSON\u3002<\/p>\n

        \u5b9a\u4f4d：\u67e5\u5382\u5546\u7684 schema \u9650\u5236\u3002strict \u5bf9\u5d4c\u5957\u6df1\u5ea6\u3001\u5355\u5bf9\u8c61\u5c5e\u6027\u6570\u91cf\u3001\u679a\u4e3e\u6570\u91cf\u90fd\u6709\u4e0a\u9650（\u4e0d\u540c\u5382\u5546\u6570\u503c\u4e0d\u540c，\u5e38\u89c1\u662f\u5d4c\u5957\u4e0d\u8d85\u8fc7\u82e5\u5e72\u5c42\u3001\u5c5e\u6027\u603b\u6570\u6709\u4e0a\u9650）\u3002<\/p>\n

        \u539f\u56e0：\u7ea6\u675f\u89e3\u7801\u7684\u5b9e\u73b0\u8981\u5728\u91c7\u6837\u671f\u7ef4\u62a4\u300c\u5f53\u524d\u5408\u6cd5 token\u300d\u7684\u72b6\u6001\u673a，schema \u592a\u590d\u6742\u4f1a\u8ba9\u72b6\u6001\u7a7a\u95f4\u7206\u70b8，\u8981\u4e48\u62a5\u9519\u8981\u4e48\u9000\u5316\u6210 best-effort\u3002<\/p>\n

        \u89e3\u51b3：\u628a\u8d85\u590d\u6742 schema \u62c6\u6210\u300c\u5148\u62bd\u9aa8\u67b6\u3001\u518d\u8865\u7ec6\u8282\u300d\u7684\u591a\u6b65\u8c03\u7528\u2014\u2014\u7b2c\u4e00\u6b65\u8ba9\u6a21\u578b\u8fd4\u56de id \u5217\u8868，\u7b2c\u4e8c\u6b65\u9010\u4e2a\u8865\u5168\u8be6\u60c5\u3002\u6216\u8005\u5728\u5e73\u9762\u5316 schema \u4e0e\u4e1a\u52a1\u53ef\u8bfb\u6027\u4e4b\u95f4\u505a\u6743\u8861，\u80fd\u5e73\u94fa\u5c31\u522b\u6df1\u5c42\u5d4c\u5957\u3002<\/p>\n

        7.3 \u8e29\u5751 3：\u628a strict \u4e0e best-effort JSON mode \u6df7\u7528<\/h4>\n

        \u73b0\u8c61：\u4ee3\u7801\u91cc\u4e00\u90e8\u5206\u8d70 strict（\u4fdd\u8bc1），\u4e00\u90e8\u5206\u8d70 response_format={"type":"json_object"} \u7684 best-effort JSON mode（\u4e0d\u4fdd\u8bc1），\u4e24\u8fb9\u884c\u4e3a\u4e0d\u4e00\u81f4，\u6392\u67e5\u65f6\u4e00\u5934\u96fe\u6c34\u3002<\/p>\n

        \u5b9a\u4f4d：grep \u5168\u5c40\u7684 response_format \/ text.format，\u53d1\u73b0\u6df7\u7528\u4e86\u4e24\u79cd\u6a21\u5f0f：\u4e00\u79cd\u5e26 strict，\u4e00\u79cd\u53ea\u8981\u6c42 json_object\u3002<\/p>\n

        \u539f\u56e0：best-effort JSON mode \u53ea\u662f\u300c\u9f13\u52b1\u300d\u6a21\u578b\u8f93\u51fa JSON，\u4e0d\u4fdd\u8bc1\u5b57\u6bb5\u548c\u7c7b\u578b；strict \u624d\u4fdd\u8bc1\u3002\u6df7\u7528\u610f\u5473\u7740\u4f60\u7684\u7cfb\u7edf\u4e00\u90e8\u5206\u6709\u786c\u4fdd\u8bc1\u3001\u4e00\u90e8\u5206\u6ca1\u6709，\u6545\u969c\u53ea\u5728\u300c\u6ca1\u6709\u300d\u7684\u90a3\u90e8\u5206\u5192\u51fa\u6765，\u5b9a\u4f4d\u6210\u672c\u9ad8\u3002<\/p>\n

        \u89e3\u51b3：\u7edf\u4e00\u8d70 strict（\u6216\u7b49\u4ef7\u7684\u5f3a\u7ea6\u675f），\u4e0d\u8981\u5728\u540c\u4e00\u7ba1\u7ebf\u91cc\u6df7\u7528\u3002\u5982\u679c\u67d0\u4e2a\u573a\u666f\u786e\u5b9e\u4e0d\u9700\u8981\u5f3a\u7ea6\u675f（\u6bd4\u5982\u5185\u90e8 debug \u5de5\u5177），\u4e5f\u8981\u663e\u5f0f\u6807\u6ce8，\u522b\u8ba9\u5b83\u548c\u5f3a\u7ea6\u675f\u8def\u5f84\u6df7\u5728\u540c\u4e00\u4e2a\u5de5\u5177\u51fd\u6570\u91cc\u3002<\/p>\n

        7.4 \u8e29\u5751 4：provider \u5347\u7ea7\u7ed5\u8fc7\u4fdd\u8bc1<\/h4>\n

        \u73b0\u8c61：\u67d0\u6b21 SDK \u6216\u6a21\u578b\u7248\u672c\u5347\u7ea7\u540e，\u539f\u672c\u7a33\u5b9a\u7684\u7ed3\u6784\u5316\u8f93\u51fa\u5f00\u59cb\u5076\u53d1\u5b57\u6bb5\u7f3a\u5931\u6216\u7c7b\u578b\u6f02\u79fb，\u4f46\u4ee3\u7801\u6ca1\u6539\u3002<\/p>\n

        \u5b9a\u4f4d：\u6bd4\u5bf9\u5347\u7ea7\u524d\u540e\u7684\u5b9e\u9645\u8fd4\u56de\u3002\u53d1\u73b0\u65b0\u7248\u672c\u7684 strict \u5b9e\u73b0\u5728\u67d0\u4e9b schema \u5f62\u6001\u4e0b\u9000\u5316（\u6216\u9ed8\u8ba4\u884c\u4e3a\u53d8\u5316），\u4fdd\u8bc1\u88ab\u6084\u6084\u524a\u5f31\u3002<\/p>\n

        \u539f\u56e0：\u5382\u5546\u7684 strict \u5b9e\u73b0\u672c\u8eab\u5728\u8fed\u4ee3，\u4e2a\u522b\u7248\u672c\u4f1a\u5f15\u5165\u56de\u5f52；\u6216\u6a21\u578b\u9ed8\u8ba4\u7248\u672c\u5207\u6362（\u6bd4\u5982 gpt-4o \u6307\u5411\u7684\u5e95\u5c42\u6a21\u578b\u53d8\u4e86），\u800c\u65b0\u6a21\u578b\u7684\u7ea6\u675f\u9075\u5faa\u5ea6\u7565\u6709\u5dee\u5f02\u3002provider \u7684\u300c\u4fdd\u8bc1\u300d\u662f\u7248\u672c\u76f8\u5173\u7684，\u4e0d\u662f\u6c38\u6052\u771f\u7406\u3002<\/p>\n

        \u89e3\u51b3：\u5ba2\u6237\u7aef\u6c38\u8fdc\u518d\u6821\u9a8c\u4e00\u904d\u3002\u5373\u4f7f provider \u627f\u8bfa strict，\u672c\u5730 Pydantic \u6821\u9a8c\u8fd9\u9053\u95f8\u4e0d\u80fd\u7701\u2014\u2014\u5b83\u662f\u5bf9 provider \u56de\u5f52\u7684\u515c\u5e95\u3002\u540c\u65f6\u9501\u6a21\u578b\u7248\u672c（\u7528\u5177\u4f53 snapshot \/ \u65e5\u671f\u7248\u672c），\u4e0d\u8981\u628a model \u5199\u6210\u4f1a\u6f02\u79fb\u7684\u522b\u540d\u3002<\/p>\n

        7.5 \u8e29\u5751 5：\u6e29\u5ea6\u6ca1\u964d，\u5076\u53d1\u683c\u5f0f\u9519<\/h4>\n

        \u73b0\u8c61：strict \u6a21\u5f0f\u4e0b\u4ecd\u7136\u5076\u53d1（\u6bd4\u5982\u5343\u5206\u4e4b\u4e00）\u8f93\u51fa\u975e\u6cd5 JSON，\u91cd\u8bd5\u5c31\u597d，\u4f46\u9ad8\u5cf0\u671f\u5806\u79ef\u3002<\/p>\n

        \u5b9a\u4f4d：\u770b\u8bf7\u6c42\u53c2\u6570，\u53d1\u73b0 temperature \u8bbe\u6210\u4e86 0.7（\u6cbf\u7528\u804a\u5929\u914d\u7f6e\u7684\u9ed8\u8ba4\u503c）\u3002<\/p>\n

        \u539f\u56e0：\u7ea6\u675f\u89e3\u7801\u901a\u5e38\u8981\u6c42 temperature \u4e3a 0 \u624d\u80fd\u7a33\u5b9a\u8d70\u5408\u6cd5\u8def\u5f84；\u6e29\u5ea6\u504f\u9ad8\u65f6，\u91c7\u6837\u4f1a\u5728\u8fb9\u754c token \u4e0a\u300c\u8bd5\u63a2\u300d\u975e\u6cd5\u5206\u652f，\u867d\u7136\u540e\u7aef\u5927\u591a\u80fd\u7ea0\u6b63，\u4f46\u5076\u53d1\u4f1a\u6f0f\u3002\u8fd9\u662f\u6700\u5bb9\u6613\u5fd8\u3001\u5374\u6700\u9ad8\u9891\u7684\u4e00\u5751\u3002<\/p>\n

        \u89e3\u51b3：\u6240\u6709\u7ed3\u6784\u5316\u8f93\u51fa\u8bf7\u6c42\u5f3a\u5236 temperature=0.0，\u5728 SDK \u5c01\u88c5\u5c42\u5199\u6b7b，\u4e0d\u5141\u8bb8\u4e1a\u52a1\u4fa7\u8986\u76d6\u3002\u628a\u5b83\u5f53\u57fa\u7840\u8bbe\u65bd\u7ea6\u675f，\u800c\u4e0d\u662f\u53ef\u9009\u9879\u3002<\/p>\n

        7.6 \u8e29\u5751 6：\u4f9d\u8d56 schema \u5374\u4e0d\u505a\u4e1a\u52a1\u6821\u9a8c，order_id \u683c\u5f0f\u9519\u7167\u6837\u8fc7<\/h4>\n

        \u73b0\u8c61：\u7cfb\u7edf\u7528 strict \u4fdd\u8bc1\u5b57\u6bb5\u9f50\u5168\u3001\u7c7b\u578b\u6b63\u786e，\u4f46\u8fd0\u8425\u53d1\u73b0\u5e93\u91cc\u51fa\u73b0\u4e00\u6279 order_id \u4e3a 12345\u3001abc \u8fd9\u7c7b\u660e\u663e\u975e\u6cd5\u7684\u503c，\u6a21\u578b\u300c\u4e25\u683c\u5730\u300d\u586b\u4e86\u9519\u683c\u5f0f\u7684\u5b57\u7b26\u4e32\u3002<\/p>\n

        \u5b9a\u4f4d：schema \u91cc order_id \u53ea\u58f0\u660e\u4e86 type:string，\u6ca1\u6709\u683c\u5f0f\u7ea6\u675f（strict \u7981\u6b62 pattern），\u4e8e\u662f\u4efb\u4f55\u5b57\u7b26\u4e32\u90fd\u5408\u6cd5\u901a\u8fc7\u3002<\/p>\n

        \u539f\u56e0：\u8fd9\u662f\u300c\u7ea6\u675f\u89e3\u7801\u7ba1\u8bed\u6cd5\u3001\u4e0d\u7ba1\u8bed\u4e49\u300d\u7684\u5178\u578b\u540e\u679c\u3002strict \u7981\u6b62 pattern \/ minLength \u8fd9\u7c7b\u7ea6\u675f，\u610f\u5473\u7740\u300c\u5b57\u7b26\u4e32\u957f\u4ec0\u4e48\u6837\u300d\u53ea\u80fd\u9760\u4e1a\u52a1\u6821\u9a8c\u3002\u53ea\u4f9d\u8d56 schema \u7684\u56e2\u961f，\u5929\u7136\u653e\u8fc7\u4e86\u6240\u6709\u300c\u7c7b\u578b\u5bf9\u4f46\u8bed\u4e49\u9519\u300d\u7684\u6570\u636e\u3002<\/p>\n

        \u89e3\u51b3：\u5728\u5ba2\u6237\u7aef\u52a0\u4e1a\u52a1\u89c4\u5219\u6821\u9a8c\u5c42（\u89c1 5.5）：order_id \u5fc5\u987b\u5339\u914d ^ORD-\\\\d{8}-\\\\d{4}$\u3001quantity>0\u3001priority \u4e0e requires_review \u7684\u4e00\u81f4\u6027\u7b49\u3002schema \u662f\u683c\u5f0f\u95f8，\u4e1a\u52a1\u6821\u9a8c\u662f\u8bed\u4e49\u95f8，\u4e24\u9053\u90fd\u8981\u3002<\/p>\n

        \u516b\u3001\u4f18\u5316\u65b9\u6848\u4e0e\u8fdb\u9636\u6253\u6cd5<\/h3>\n

        8.1 Schema \u8bbe\u8ba1\u7eaa\u5f8b<\/h4>\n

        \u628a 3.4 \u548c\u7b2c\u4e03\u7ae0\u7684\u6559\u8bad\u6536\u6210\u51e0\u6761\u53ef\u6267\u884c\u7684\u7eaa\u5f8b：<\/p>\n

          \n
        • required \u8d8a\u5c11\u8d8a\u597d：\u6bcf\u4e00\u4e2a required \u90fd\u662f\u6f5c\u5728\u5931\u8d25\u70b9\u3002\u80fd\u9ed8\u8ba4\u3001\u80fd\u63a8\u65ad\u7684\u5c31\u4e0d\u8981\u585e\u8fdb required\u3002strict \u4e0b required \u662f\u5f3a\u5236\u7684，\u6240\u4ee5\u300c\u53ef\u9009\u300d\u8981\u7528 ["type","null"] \u8868\u8fbe，\u800c\u4e0d\u662f\u7701\u7565\u3002<\/li>\n
        • \u663e\u5f0f\u7c7b\u578b：price \u662f number \u4e0d\u662f\u300c\u542b\u8d27\u5e01\u7b26\u7684\u5b57\u7b26\u4e32\u300d；quantity \u662f integer \u4e0d\u662f string\u3002\u7c7b\u578b\u9519\u4e86\u4e0b\u6e38\u5fc5\u5d29\u3002<\/li>\n
        • \u5c01\u95ed\u9009\u62e9\u7528 enum：\u4f18\u5148\u7ea7\u3001\u72b6\u6001\u8fd9\u7c7b\u6709\u9650\u96c6\u5408\u7528 enum，\u522b\u7528\u81ea\u7531 string \u8ba9\u6a21\u578b\u81ea\u5df1\u7f16\u8bcd\u3002<\/li>\n
        • \u907f\u514d\u5bbd string \u5b57\u6bb5：\u4e00\u4e2a description: string \u88c5\u6240\u6709\u4fe1\u606f\u7684\u8bbe\u8ba1，\u7b49\u4e8e\u628a\u7ed3\u6784\u5316\u8f93\u51fa\u9000\u5316\u6210\u975e\u7ed3\u6784\u5316\u3002\u8be5\u62c6\u5b57\u6bb5\u5c31\u62c6\u3002<\/li>\n
        • description \u4e0d\u66ff\u4ee3\u7ea6\u675f：description \u662f\u7ed9\u6a21\u578b\u770b\u7684\u63d0\u793a，\u4e0d\u662f\u7ed9\u89e3\u6790\u5668\u770b\u7684\u4fdd\u8bc1\u3002\u80fd\u7528\u7c7b\u578b\/enum\/required \u8868\u8fbe\u7684，\u522b\u4f9d\u8d56 description \u8ba9\u6a21\u578b\u300c\u81ea\u89c9\u300d\u3002<\/li>\n<\/ul>\n

          8.2 \u91cd\u8bd5\u4e0e\u7f13\u5b58\u7684\u5de5\u7a0b\u53d6\u820d<\/h4>\n

          \u91cd\u8bd5：\u5355\u6b21\u300c\u7d27\u5316\u91cd\u8bd5\u300d\u5373\u53ef\u2014\u2014\u628a\u4e0a\u4e00\u6b21\u7684\u6821\u9a8c\u9519\u8bef\u4f5c\u4e3a feedback \u5582\u56de\u6a21\u578b（5.5 \u7684 feedback \u53c2\u6570），\u8ba9\u5b83\u77e5\u9053\u54ea\u91cc\u9519\u4e86\u3002\u4e0d\u8981\u65e0\u9650\u91cd\u8bd5：\u8d85\u8fc7\u4e00\u6b21\u5c31\u964d\u7ea7\u4eba\u5de5\u961f\u5217，\u5426\u5219\u9ad8\u5cf0\u671f\u4f1a\u5361\u6b7b\u5728\u53cd\u590d\u751f\u6210\u4e0a\u3002<\/p>\n

          \u8bed\u4e49\u7f13\u5b58：\u7ed3\u6784\u5316\u8f93\u51fa\u5e38\u914d\u5408\u8bed\u4e49\u7f13\u5b58\u7701 token\u3002\u4f46\u7f13\u5b58\u7684\u76f8\u4f3c\u5ea6\u9608\u503c\u4e0d\u80fd\u8c03\u592a\u9ad8\u2014\u2014\u9608\u503c\u8fc7\u9ad8\u4f1a\u628a\u300c\u8bed\u4e49\u4e0d\u540c\u4f46\u8868\u9762\u76f8\u4f3c\u300d\u7684\u8bf7\u6c42\u5f53\u6210\u540c\u4e00\u9898，\u8fd4\u56de\u7b54\u975e\u6240\u95ee\u7684\u7ed3\u6784\u5316\u7ed3\u679c\u3002\u9608\u503c\u8bbe\u5b9a\u8981\u57fa\u4e8e\u4f60\u7684\u4e1a\u52a1\u5bb9\u5fcd\u5ea6\u505a A\/B，\u522b\u62cd\u8111\u888b\u3002<\/p>\n

          8.3 \u591a\u5382\u5546 schema \u7684\u7ffb\u8bd1\u5c42<\/h4>\n

          \u5982\u679c\u4f60\u7684\u7cfb\u7edf\u8981\u540c\u65f6\u63a5 OpenAI \/ Anthropic \/ Gemini，\u4e0d\u8981\u4e3a\u6bcf\u5bb6\u5199\u4e00\u4efd schema\u3002\u7528\u4e00\u4e2a\u4e2d\u7acb\u8868\u793a（Pydantic \u662f\u5929\u7136\u9009\u62e9），\u5bfc\u51fa JSON schema \u540e，\u6309\u5404\u5382\u5546\u7684\u7eaa\u5f8b\u505a\u8f7b\u91cf\u7ffb\u8bd1（\u8865 additionalProperties\u3001\u5904\u7406 optional \u5199\u6cd5\u3001\u52a0 propertyOrdering）\u3002\u7ffb\u8bd1\u5c42\u96c6\u4e2d\u7ef4\u62a4，\u907f\u514d\u4e09\u5bb6\u884c\u4e3a\u6f02\u79fb\u3002<\/p>\n

          \u4e5d\u3001\u9002\u7528\u573a\u666f\u4e0e\u9009\u578b\u5efa\u8bae<\/h3>\n

          9.1 \u4ec0\u4e48\u60c5\u51b5\u5fc5\u987b\u4e0a\u7ed3\u6784\u5316\u8f93\u51fa<\/h4>\n
            \n
          • \u6a21\u578b\u8f93\u51fa\u8981\u76f4\u63a5\u8fdb\u6570\u636e\u5e93\u3001\u8fdb\u4e0b\u6e38 API\u3001\u8fdb\u62a5\u8868\u2014\u2014\u4efb\u4f55\u300c\u4e0b\u6e38\u662f\u673a\u5668\u300d\u7684\u573a\u666f\u3002<\/li>\n
          • \u8f93\u51fa\u8981\u88ab\u4ee3\u7801\u7a33\u5b9a\u89e3\u6790，\u4e0d\u80fd\u5bb9\u5fcd\u5076\u53d1\u683c\u5f0f\u9519\u3002<\/li>\n
          • \u591a\u6b65\u9aa4 Agent \u4e4b\u95f4\u8981\u7528\u7ed3\u6784\u5316\u6d88\u606f\u4f20\u9012\u72b6\u6001\u3002<\/li>\n<\/ul>\n

            9.2 \u4ec0\u4e48\u60c5\u51b5\u4e0d\u8be5\u8fd9\u4e48\u4e0a<\/h4>\n

            \u4e0d\u8be5\u53ea\u7528 prompt \u7ea6\u675f：\u751f\u4ea7\u73af\u5883\u91cc\u300c\u8bf7\u8fd4\u56de JSON\u300d\u4e0d\u591f，\u5fc5\u987b\u4e0a strict \u6216\u7b49\u4ef7\u7ea6\u675f\u89e3\u7801\u3002<\/p>\n

            \u4e0d\u8be5\u6df7\u7528 strict \u4e0e best-effort：\u540c\u4e00\u7ba1\u7ebf\u884c\u4e3a\u4e0d\u4e00\u81f4，\u6545\u969c\u53ea\u5728\u5f31\u7ea6\u675f\u90a3\u4fa7\u5192\u5934\u3002<\/p>\n

            \u4e0d\u8be5\u8df3\u8fc7\u4e1a\u52a1\u6821\u9a8c：schema \u7ba1\u8bed\u6cd5\u4e0d\u7ba1\u8bed\u4e49，\u4f9d\u8d56\u5b83\u653e\u8fc7\u4e86\u6240\u6709\u300c\u7c7b\u578b\u5bf9\u4f46\u8bed\u4e49\u9519\u300d\u7684\u810f\u6570\u636e\u3002<\/p>\n

            \u4e0d\u8be5\u65e0\u9650\u91cd\u8bd5：\u5355\u6b21\u7d27\u5316\u91cd\u8bd5\u540e\u76f4\u63a5\u964d\u7ea7\u4eba\u5de5，\u907f\u514d\u9ad8\u5cf0\u671f\u5361\u6b7b\u3002<\/p>\n

            \u4e0d\u8be5\u628a\u8bed\u4e49\u7f13\u5b58\u9608\u503c\u8c03\u592a\u9ad8：\u4f1a\u7b54\u975e\u6240\u95ee，\u4e14\u9519\u8bef\u662f\u7ed3\u6784\u5316\u7684（\u66f4\u96be\u88ab\u53d1\u73b0）\u3002<\/p>\n

            9.3 \u4e00\u53e5\u8bdd\u51b3\u7b56<\/h4>\n

            \u4e0b\u6e38\u662f\u673a\u5668？
            \n+– \u662f -> \u9700\u8981\u5de5\u5177\u52a8\u4f5c？
            \n +– \u662f -> \u5f00 tool calling（strict tool schema）
            \n +– \u5426 -> \u5f00 structured output（strict \/ output_config \/ response_json_schema）
            \n \u4e24\u8005\u90fd\u8981？\u5148 tool calling \u518d structured output \u6536\u5c3e
            \n+– \u5426 -> \u7eaf\u804a\u5929，\u968f\u610f
            \n\u94c1\u5f8b: temperature=0 + additionalProperties:false + \u5ba2\u6237\u7aef Pydantic \u518d\u9a8c + \u4e1a\u52a1\u89c4\u5219\u6821\u9a8c<\/p>\n

            \u5341\u3001\u603b\u7ed3\u4e0e\u5c55\u671b<\/h3>\n

            \u628a\u5168\u6587\u7684\u8d26\u91cd\u7b97\u4e00\u904d，\u7ed3\u8bba\u662f\u4e94\u6761：<\/p>\n

            \u7b2c\u4e00，\u7ed3\u6784\u5316\u8f93\u51fa\u662f\u7cfb\u7edf\u5de5\u7a0b，\u4e0d\u662f prompt \u5de5\u7a0b\u3002\u4e00\u53e5\u300c\u8bf7\u8fd4\u56de JSON\u300d\u5728 demo \u91cc\u80fd\u8dd1，\u5728\u751f\u4ea7\u91cc\u662f 3% \u7684\u810f\u6570\u636e\u7387，\u662f\u8d22\u52a1\u5bf9\u8d26\u5d29\u6e83\u3002\u5fc5\u987b\u4e0a\u534f\u8bae\u7ea7\u7ea6\u675f\u3002<\/p>\n

            \u7b2c\u4e8c，\u4e24\u6761\u539f\u8bed\u6b63\u4ea4：tool calling \u662f\u4e2d\u95f4\u52a8\u4f5c，structured output \u662f\u6700\u7ec8\u7b54\u6848\u3002\u7ec4\u5408\u4f7f\u7528\u65f6\u5148\u52a8\u4f5c\u540e\u7ed3\u8bba，\u522b\u6df7\u6dc6\u3002<\/p>\n

            \u7b2c\u4e09，\u7ea6\u675f\u89e3\u7801\u548c\u4e8b\u540e\u6821\u9a8c\u5fc5\u987b\u53e0\u52a0\u3002\u524d\u8005\u7ba1\u8bed\u6cd5（\u683c\u5f0f\u3001\u7c7b\u578b\u3001\u5b57\u6bb5），\u540e\u8005\u7ba1\u8bed\u4e49（quantity>0\u3001order_id \u683c\u5f0f\u3001\u4e1a\u52a1\u4e00\u81f4\u6027）\u3002\u4efb\u4f55\u4e00\u9053\u7f3a\u5931\u90fd\u4f1a\u653e\u8fc7\u9519\u7c7b\u4e8b\u6545\u3002<\/p>\n

            \u7b2c\u56db，provider \u7684 strict \u4fdd\u8bc1\u662f\u7248\u672c\u76f8\u5173\u7684，\u5ba2\u6237\u7aef\u6821\u9a8c\u4e0d\u80fd\u7701\u3002\u6e29\u5ea6\u6ca1\u964d\u3001additionalProperties \u6f0f\u5199\u3001strict \u4e0e best-effort \u6df7\u7528\u3001provider \u5347\u7ea7\u7ed5\u8fc7\u4fdd\u8bc1\u2014\u2014\u8fd9\u56db\u7c7b\u5751\u90fd\u9760\u300c\u5ba2\u6237\u7aef\u518d\u9a8c\u4e00\u904d\u300d\u515c\u5e95\u3002<\/p>\n

            \u7b2c\u4e94，schema \u8bbe\u8ba1\u672c\u8eab\u5c31\u662f\u53ef\u9760\u6027\u7684\u4e00\u90e8\u5206\u3002required \u8d8a\u5c11\u8d8a\u597d\u3001\u7c7b\u578b\u663e\u5f0f\u3001\u5c01\u95ed\u9009\u62e9\u7528 enum\u3001description \u4e0d\u66ff\u4ee3\u7ea6\u675f\u3002\u70c2 schema \u518d\u5f3a\u7684\u7ea6\u675f\u4e5f\u6551\u4e0d\u56de\u6765\u3002<\/p>\n

            \u5f80\u524d\u770b，\u4e24\u4ef6\u4e8b\u503c\u5f97\u5173\u6ce8\u3002\u4e00\u662f BFCL \u8fd9\u7c7b\u300c\u7ed3\u6784\u5316\u6d88\u8d39\u300d\u57fa\u51c6\u4f1a\u8d8a\u6765\u8d8a\u91cd\u8981，\u6a21\u578b\u5728\u804a\u5929\u699c\u4e0a\u7684\u5206\u6570\u4f1a\u8d8a\u6765\u8d8a\u591a\u5730\u88ab\u300c\u80fd\u4e0d\u80fd\u6b63\u786e\u586b\u53c2\u6570\u300d\u7a00\u91ca\u3002\u4e8c\u662f \u7ea6\u675f\u89e3\u7801\u4f1a\u4e0b\u6c89\u5230\u66f4\u591a\u81ea\u6258\u7ba1\u6846\u67b6，XGrammar \u8fd9\u7c7b\u9ad8\u6027\u80fd\u5b9e\u73b0\u8ba9\u672c\u5730\u6a21\u578b\u7684\u4e25\u683c\u7ed3\u6784\u5316\u8f93\u51fa\u51e0\u4e4e\u96f6\u6210\u672c，\u95ed\u6e90 strict \u4e0d\u518d\u662f\u62a4\u57ce\u6cb3\u3002\u4f46\u65e0\u8bba\u673a\u5236\u600e\u4e48\u6f14\u8fdb，\u300c\u5ba2\u6237\u7aef\u6821\u9a8c + \u4e1a\u52a1\u89c4\u5219 + temperature=0\u300d\u8fd9\u5957\u94c1\u5f8b\u4e0d\u4f1a\u53d8\u3002<\/p>\n

            \u53c2\u8003\u7d20\u6750\u4e0e\u5ef6\u4f38\u9605\u8bfb<\/h3>\n
              \n
            • Structured Output: \u5404\u5382\u5546\u80fd\u529b\u5bf9\u7167（OpenAI \/ Anthropic \/ Gemini） \u2014 AI Wiki，2026-09<\/li>\n
            • How to Get Structured JSON Output from LLMs (2026 Guide): \u9010\u5382\u5546\u53c2\u6570 \u2014 Andrew \u7684\u6307\u5357，2026-09<\/li>\n
            • Structured Outputs and Constrained Decoding: \u7ea6\u675f\u89e3\u7801 vs \u4e8b\u540e\u6821\u9a8c，\u8ba2\u5355 schema \u5b9e\u4f8b \u2014 Technology on the Net，2026-09<\/li>\n
            • Reliable Tool Calling: \u4e24\u79cd\u539f\u8bed\u6b63\u4ea4\u3001BFCL \u57fa\u51c6\u3001\u5e38\u89c1\u6545\u969c\u6a21\u5f0f \u2014 ChangeGamer AI，2026-09<\/li>\n
            • OpenAI Structured Outputs \u5b98\u65b9\u6587\u6863（strict \/ json_schema） \u2014 OpenAI，2026-10<\/li>\n
            • Anthropic Messages API: output_config（GA） \u2014 Anthropic，2026-10<\/li>\n
            • Gemini API: response_json_schema \u4e0e propertyOrdering \u2014 Google，2026-10<\/li>\n
            • Outlines: \u7ea6\u675f\u89e3\u7801 Python \u5e93 \u2014 dottxt，2026<\/li>\n
            • XGrammar: \u9ad8\u6027\u80fd\u7ea6\u675f\u89e3\u7801\u5f15\u64ce \u2014 xgrammar-project，2026<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"

              \u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316：\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 \u672c\u6587\u628a\u300c\u8ba9\u6a21\u578b\u8f93\u51fa\u80fd\u88ab\u4ee3\u7801\u53ef\u9760\u6d88\u8d39\u300d\u62c6\u6210\u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed\u2014\u2014tool\/function calling（\u4e2d\u95f4\u52a8\u4f5c）\u4e0e structured outputs\/JSON mode（\u6700\u7ec8\u7b54\u6848），\u8bb2\u6e05\u4e3a\u4ec0\u4e48\u751f\u4ea7\u73af\u5883\u91cc\u4e00\u53e5\u300c\u8bf7\u8fd4\u56de JSON\u300d\u8fdc\u8fdc\u4e0d\u591f\u3002\u6838\u5fc3\u7ae0\u8282\u7ed9\u51fa 2026-09 \u4e09\u5927\u5382\u5546（OpenAI \/ <\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[9334,13948,13947],"topic":[],"class_list":["post-121871","post","type-post","status-publish","format-standard","hentry","category-server","tag-tool","tag-13948","tag-13947"],"yoast_head":"\n\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/121871.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316：\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 \u672c\u6587\u628a\u300c\u8ba9\u6a21\u578b\u8f93\u51fa\u80fd\u88ab\u4ee3\u7801\u53ef\u9760\u6d88\u8d39\u300d\u62c6\u6210\u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed\u2014\u2014tool\/function calling（\u4e2d\u95f4\u52a8\u4f5c）\u4e0e structured outputs\/JSON mode（\u6700\u7ec8\u7b54\u6848），\u8bb2\u6e05\u4e3a\u4ec0\u4e48\u751f\u4ea7\u73af\u5883\u91cc\u4e00\u53e5\u300c\u8bf7\u8fd4\u56de JSON\u300d\u8fdc\u8fdc\u4e0d\u591f\u3002\u6838\u5fc3\u7ae0\u8282\u7ed9\u51fa 2026-09 \u4e09\u5927\u5382\u5546（OpenAI \/\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/121871.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-03T14:24:43+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121871.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/121871.html\",\"name\":\"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-11-03T14:24:43+00:00\",\"dateModified\":\"2026-11-03T14:24:43+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/121871.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/121871.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/121871.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/121871.html","og_locale":"zh_CN","og_type":"article","og_title":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316：\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 \u672c\u6587\u628a\u300c\u8ba9\u6a21\u578b\u8f93\u51fa\u80fd\u88ab\u4ee3\u7801\u53ef\u9760\u6d88\u8d39\u300d\u62c6\u6210\u4e24\u6761\u6b63\u4ea4\u7684\u539f\u8bed\u2014\u2014tool\/function calling（\u4e2d\u95f4\u52a8\u4f5c）\u4e0e structured outputs\/JSON mode（\u6700\u7ec8\u7b54\u6848），\u8bb2\u6e05\u4e3a\u4ec0\u4e48\u751f\u4ea7\u73af\u5883\u91cc\u4e00\u53e5\u300c\u8bf7\u8fd4\u56de JSON\u300d\u8fdc\u8fdc\u4e0d\u591f\u3002\u6838\u5fc3\u7ae0\u8282\u7ed9\u51fa 2026-09 \u4e09\u5927\u5382\u5546（OpenAI \/","og_url":"https:\/\/www.wsisp.com\/helps\/121871.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-11-03T14:24:43+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"12 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/121871.html","url":"https:\/\/www.wsisp.com\/helps\/121871.html","name":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-11-03T14:24:43+00:00","dateModified":"2026-11-03T14:24:43+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/121871.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/121871.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/121871.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u7ed3\u6784\u5316\u8f93\u51fa\u4e0e Function Calling \u5de5\u7a0b\u5316\uff1a\u8ba9\u6a21\u578b\u8f93\u51fa\u53ef\u88ab\u673a\u5668\u53ef\u9760\u6d88\u8d39"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121871","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=121871"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/121871\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=121871"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=121871"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=121871"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=121871"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}