{"id":103505,"date":"2026-09-10T19:17:40","date_gmt":"2026-09-10T11:17:40","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/103505.html"},"modified":"2026-09-10T19:17:40","modified_gmt":"2026-09-10T11:17:40","slug":"glm-5-3-flash-%e6%88%90%e6%9c%ac%e4%bc%98%e5%8c%96%e5%ae%9e%e6%88%98%ef%bc%9a%e5%8a%a0%e4%b8%80%e4%b8%aa%e5%8f%82%e6%95%b0%ef%bc%8c%e6%8e%a8%e7%90%86%e6%88%90%e6%9c%ac%e7%9b%b4%e9%99%8d-6-3-%e5%80%8d","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/103505.html","title":{"rendered":"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d"},"content":{"rendered":"

\u6211\u4eec\u95ee\u4e86 GLM-5.3-Flash \u4e00\u4e2a\u975e\u5e38\u7b80\u5355\u7684\u95ee\u9898：SSH \u9ed8\u8ba4\u4f7f\u7528\u54ea\u4e2a\u7aef\u53e3\u3002\u5b83\u7528\u4e86 625 \u4e2a\u8f93\u51fa Token，\u6700\u540e\u624d\u7ed9\u51fa\u7b54\u6848：\u201c22\u201d\u3002<\/p>\n

\u8fd9\u4e0d\u662f\u6211\u4eec\u7279\u610f\u6311\u51fa\u6765\u7684\u6781\u7aef\u6848\u4f8b，\u800c\u662f 5 \u6b21\u6d4b\u8bd5\u7684\u4e2d\u4f4d\u6570\u3002Qwen3.8-Max \u56de\u7b54\u540c\u4e00\u4e2a\u95ee\u9898\u53ea\u7528\u4e86 156 \u4e2a Token\u3002Flash \u8fd9\u4e9b\u8c03\u7528\u4e2d，\u6211\u4eec\u652f\u4ed8\u7684\u5927\u90e8\u5206\u8f93\u51fa\u6210\u672c，\u5176\u5b9e\u90fd\u82b1\u5728\u4e86\u7528\u6237\u6839\u672c\u770b\u4e0d\u5230\u7684\u63a8\u7406\u8fc7\u7a0b\u4e0a，\u800c\u8fd9\u4e2a\u95ee\u9898\u6700\u7ec8\u7684\u7b54\u6848\u53ea\u6709\u4e24\u4e2a\u5b57\u7b26\u3002<\/p>\n

\u7ed3\u679c\u53d1\u73b0，\u89e3\u51b3\u529e\u6cd5\u53ea\u9700\u8981\u6539\u4e00\u884c\u53c2\u6570\u3002\u5728\u8bf7\u6c42\u91cc\u52a0\u5165 reasoning_effort: "low"，\u540c\u4e00\u4e2a\u95ee\u9898\u5c31\u53ea\u9700\u8981 155 \u4e2a Token\u3002GLM-5.3-Flash \u9ed8\u8ba4\u4f1a\u4f7f\u7528\u6700\u9ad8\u7684\u63a8\u7406\u5f3a\u5ea6，\u4f46\u4ec5\u4ece\u8bf7\u6c42\u672c\u8eab\u4f60\u6839\u672c\u770b\u4e0d\u51fa\u6765\u3002\u53ea\u589e\u52a0\u8fd9\u4e00\u4e2a\u53c2\u6570，\u5c31\u8ba9 Flash \u5728\u6211\u4eec\u7684\u6574\u5957 Prompt \u6d4b\u8bd5\u4e2d\u4fbf\u5b9c\u4e86 6.3 \u500d\u3002<\/p>\n

\u8fd9\u4e00\u70b9\u5bf9\u5927\u591a\u6570\u4eba\u6765\u8bb2\u5f88\u91cd\u8981，\u56e0\u4e3a\u6a21\u578b\u662f\u6309 Token \u6536\u8d39\u7684，\u4f46\u4ef7\u683c\u9875\u9762\u4e0d\u4f1a\u544a\u8bc9\u4f60\u4e00\u4e2a\u6a21\u578b\u901a\u5e38\u4f1a\u751f\u6210\u591a\u5c11 Token\u3002GLM-5.3-Flash \u7684\u8f93\u51fa\u4ef7\u683c\u53ea\u6709 Qwen3.8-Max \u7684\u5341\u4e8c\u5206\u4e4b\u4e00，\u4e5f\u53ea\u6709 Kimi K3 \u7684\u4e09\u5341\u5206\u4e4b\u4e00\u3002\u4e3a\u4e86\u770b\u770b\u8fd9\u4e9b\u7eb8\u9762\u4ef7\u683c\u6bd4\u4f8b\u843d\u5230\u771f\u5b9e\u8bf7\u6c42\u91cc\u5230\u5e95\u662f\u4ec0\u4e48\u6837，\u6211\u4eec\u5728 DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u4e0a\u5b9e\u9645\u8dd1\u4e86 2700 \u6b21 API \u8c03\u7528\u3002<\/p>\n

\u4f7f\u7528\u9ed8\u8ba4\u914d\u7f6e\u65f6，Flash \u76f8\u6bd4 Qwen \u7684\u6210\u672c\u4f18\u52bf\u662f 2.3 \u500d，\u800c\u4e0d\u662f 12 \u500d\u3002\u663e\u5f0f\u914d\u7f6e\u540e，\u8fd9\u4e2a\u4f18\u52bf\u53d8\u6210\u4e86 14.5 \u500d；\u76f8\u6bd4 Kimi K3，\u5219\u8fbe\u5230 106 \u500d\u3002\u8fd9\u51e0\u4e2a\u6570\u5b57\u4e4b\u95f4\u7684\u5dee\u8ddd，\u53ea\u6765\u81ea\u4e00\u4e2a\u53c2\u6570\u3002\u672c\u6587\u540e\u9762\u4f1a\u5b8c\u6574\u8bb2\u6e05\u695a\u6211\u4eec\u662f\u600e\u4e48\u627e\u5230\u5b83\u7684，\u5305\u62ec\u4e00\u5f00\u59cb\u8d70\u9519\u7684\u90a3\u6761\u8def\u3002<\/p>\n

\u6574\u4e2a\u6d4b\u8bd5\u603b\u5171\u82b1\u4e86 14.75 \u7f8e\u5143\u3002\u4e0b\u9762\u6240\u6709\u7ed3\u679c\u90fd\u53ef\u4ee5\u901a\u8fc7\u8fd9\u4e2a\u4ee3\u7801\u4ed3\u5e93(github.com\/Jameshskelton\/GLM-Flash-Analysis)\u590d\u73b0，\u5305\u62ec Prompt\u3001\u6d4b\u8bd5\u811a\u672c\u3001\u539f\u59cb\u65e5\u5fd7，\u4ee5\u53ca\u91cd\u65b0\u751f\u6210\u6587\u4e2d\u6240\u6709\u56fe\u8868\u7684\u811a\u672c\u3002<\/p>\n

GLM-5.3-Flash \u662f\u4ec0\u4e48：3200 \u4ebf\u53c2\u6570，\u6bcf\u4e2a Token \u6fc0\u6d3b 180 \u4ebf，\u91c7\u7528 MIT \u8bb8\u53ef\u8bc1<\/h3>\n

\"images<\/p>\n

GLM-5.3-Flash \u662f Z.ai GLM-5 \u7cfb\u5217\u4e2d\u9996\u4e2a\u539f\u751f\u591a\u6a21\u6001\u6a21\u578b，\u91c7\u7528 MIT \u8bb8\u53ef\u8bc1\u53d1\u5e03\u3002\u5b83\u5e76\u4e0d\u662f\u5728 GLM-5.2 \u57fa\u7840\u4e0a\u505a\u7684\u4e00\u6b21\u540e\u8bad\u7ec3\u66f4\u65b0，\u800c\u662f\u4ece\u4e00\u4e2a\u91cd\u65b0\u8bad\u7ec3\u7684\u57fa\u7840\u6a21\u578b\u5f00\u59cb，\u6574\u4e2a\u67b6\u6784\u548c\u8bad\u7ec3\u65b9\u6848\u90fd\u56f4\u7ed5\u6548\u7387\u8fdb\u884c\u4e86\u91cd\u65b0\u8bbe\u8ba1\u3002<\/p>\n

\u6700\u503c\u5f97\u5173\u6ce8\u7684\u67b6\u6784\u6570\u5b57，\u4e5f\u662f\u548c\u6210\u672c\u6700\u76f8\u5173\u7684\u6570\u5b57，\u662f：\u603b\u53c2\u6570\u91cf 3200 \u4ebf，\u4f46\u6bcf\u4e2a Token \u53ea\u6fc0\u6d3b 180 \u4ebf\u53c2\u6570\u3002 Z.ai \u8868\u793a，GLM-5.3-Flash \u5728\u591a\u9879 Benchmark \u4e0a\u8d85\u8fc7 GLM-5.2，\u800c\u4ef7\u683c\u53ea\u6709\u540e\u8005\u7684\u5341\u5206\u4e4b\u4e00；\u540c\u65f6\u5728\u4ee3\u7801\u548c Agent \u4efb\u52a1\u4e0a\u63a5\u8fd1 Claude Opus 4.8，\u4f8b\u5982 Terminal-Bench 2.1 \u5f97\u5206 84.3，DeepSWE \u5f97\u5206 63.4\u3002<\/p>\n

\u8fd9\u4e3b\u8981\u7531\u4e24\u9879\u67b6\u6784\u53d8\u5316\u63a8\u52a8\u3002\u7b2c\u4e00\u9879\u662f\u5c06\u7a00\u758f\u6ce8\u610f\u529b\u4e0e\u7ebf\u6027\u6ce8\u610f\u529b\u7ed3\u5408\u8d77\u6765\u7684\u6df7\u5408\u6ce8\u610f\u529b\u8bbe\u8ba1，\u8fd9\u662f GLM \u7cfb\u5217\u9996\u6b21\u91c7\u7528\u8fd9\u79cd\u67b6\u6784，\u76ee\u6807\u662f\u5728\u4fdd\u7559\u957f\u4e0a\u4e0b\u6587\u80fd\u529b\u7684\u540c\u65f6，\u964d\u4f4e\u957f\u4e0a\u4e0b\u6587\u63a8\u7406\u6210\u672c\u3002\u7b2c\u4e8c\u9879\u662f Manifold-Constrained Hyper-Connections，Z.ai \u7528\u5b83\u6765\u63d0\u5347\u6a21\u578b\u6269\u5c55\u6548\u7387\u3002\u8fd9\u4e24\u9879\u8bbe\u8ba1\u90fd\u5efa\u7acb\u5728\u4e00\u4e2a\u5305\u542b 30 \u4e07\u4ebf Token \u7684\u591a\u6a21\u6001\u9884\u8bad\u7ec3\u8bed\u6599\u4e4b\u4e0a\u3002<\/p>\n

\u5982\u679c\u628a\u5b83\u548c DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406 \u6700\u8fd1\u4e0a\u7ebf\u7684\u5176\u4ed6\u5f00\u653e\u6743\u91cd\u6a21\u578b\u653e\u5728\u4e00\u8d77\u770b，\u6700\u6709\u610f\u601d\u7684\u662f\u5b83\u7684\u8ba1\u7b97\u89c4\u6a21\u3002Qwen3.8-Max \u662f\u4e00\u4e2a 2.4 \u4e07\u4ebf\u53c2\u6570\u7684\u6a21\u578b，\u6bcf\u4e2a Token \u6fc0\u6d3b 950 \u4ebf\u53c2\u6570；Kimi K3 \u7684\u89c4\u6a21\u66f4\u5927\u3002\u800c Flash \u6bcf\u4e2a Token \u53ea\u6fc0\u6d3b 180 \u4ebf\u53c2\u6570\u2014\u2014\u5927\u7ea6\u53ea\u6709 Qwen \u6fc0\u6d3b\u53c2\u6570\u91cf\u7684\u4e94\u5206\u4e4b\u4e00，\u76f8\u6bd4 Kimi \u5219\u66f4\u4f4e\u5f97\u591a\u2014\u2014\u4f46\u5b83\u5728 Artificial Analysis Intelligence Index \u4e0a\u7684\u5f97\u5206\u51e0\u4e4e\u548c Kimi K3 Max \u4e00\u6837\u3002<\/p>\n

\u8fd9\u5c31\u662f GLM-5.3-Flash \u7684\u6838\u5fc3\u5356\u70b9：\u4ee5\u8fdc\u4f4e\u4e8e\u4f20\u7edf\u5927\u6a21\u578b\u7684\u5355 Token \u8ba1\u7b97\u91cf，\u63d0\u4f9b\u63a5\u8fd1\u524d\u6cbf\u6a21\u578b\u7684\u80fd\u529b\u3002\u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48 DigitalOcean \u53ef\u4ee5\u628a\u5b83\u7684\u8f93\u5165\u4ef7\u683c\u5b9a\u5728 0.15 \u7f8e\u5143 \/ \u767e\u4e07 Token，\u800c Kimi \u5219\u662f 3.00 \u7f8e\u5143 \/ \u767e\u4e07 Token\u3002<\/p>\n

\u8fd9\u4e2a\u5356\u70b9\u672c\u8eab\u786e\u5b9e\u6210\u7acb，\u5bf9\u5e94 Benchmark \u6570\u636e\u4e5f\u6765\u81ea Z.ai \u5b98\u65b9\u3002\u4f46\u6709\u4e00\u4e2a\u6570\u5b57\u51e0\u4e4e\u6ca1\u4eba\u4f1a\u516c\u5e03：\u4e00\u4e2a\u6a21\u578b\u4e3a\u4e86\u5f97\u5230\u7b54\u6848，\u5230\u5e95\u4f1a\u4f7f\u7528\u591a\u5c11 Token\u3002\u800c\u6211\u4eec\u6700\u7ec8\u53d1\u73b0，\u76f8\u5f53\u4e00\u90e8\u5206\u7eb8\u9762\u6210\u672c\u4f18\u52bf，\u6070\u6070\u6d88\u8017\u5728\u8fd9\u91cc\u3002<\/p>\n

\u6211\u4eec\u5982\u4f55\u6d4b\u91cf DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u4e0a\u7684\u5b9e\u9645\u6210\u672c<\/h3>\n

DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u4e0a\u6709\u4e09\u4e2a\u5f00\u653e\u6743\u91cd\u6a21\u578b：glm-5.3-flash\u3001qwen3.8-max \u548c kimi-k3\u3002\u5b83\u4eec\u5728 DigitalOcean \u4e0a\u6bcf\u767e\u4e07 Token \u7684\u4ef7\u683c\u5982\u4e0b：<\/p>\n\n\u6a21\u578b\u8f93\u5165\u8f93\u51fa\u7f13\u5b58\u8f93\u5165<\/tr>\n\n\n\n\n
glm-5.3-flash<\/td>\n0.15 \u7f8e\u5143<\/td>\n0.50 \u7f8e\u5143<\/td>\n0.03 \u7f8e\u5143<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\n2.00 \u7f8e\u5143<\/td>\n6.00 \u7f8e\u5143<\/td>\n0.20 \u7f8e\u5143<\/td>\n<\/tr>\n
kimi-k3<\/td>\n3.00 \u7f8e\u5143<\/td>\n15.00 \u7f8e\u5143<\/td>\n0.30 \u7f8e\u5143<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u6211\u4eec\u51c6\u5907\u4e86\u4e24\u7ec4 Prompt\u3002\u7b2c\u4e00\u7ec4\u5305\u542b 24 \u4e2a\u5f00\u653e\u5f0f\u95ee\u9898，\u5206\u4e3a\u4e09\u4e2a\u96be\u5ea6\u5c42\u7ea7：Tier 1 \u662f\u7c7b\u4f3c\u201cSSH \u9ed8\u8ba4\u7aef\u53e3\u662f\u4ec0\u4e48\u201d\u8fd9\u6837\u7684\u7b80\u5355\u67e5\u8be2\u95ee\u9898；Tier 2 \u662f\u4e2d\u7b49\u96be\u5ea6\u5206\u6790\u95ee\u9898，\u6bd4\u5982\u4e3a\u4ec0\u4e48\u6279\u5904\u7406\u80fd\u63d0\u9ad8\u541e\u5410\u91cf，\u5374\u4e0d\u4e00\u5b9a\u964d\u4f4e\u5355\u6b21\u8bf7\u6c42\u5ef6\u8fdf；Tier 3 \u5219\u662f\u5e26\u591a\u4e2a\u7ea6\u675f\u6761\u4ef6\u7684\u590d\u6742\u95ee\u9898，\u4f8b\u5982\u5728\u56fa\u5b9a\u9884\u7b97\u4e0b\u9009\u62e9\u90e8\u7f72\u62d3\u6251\u3002\u6240\u6709\u95ee\u9898\u90fd\u6ca1\u6709\u6307\u5b9a\u8f93\u51fa\u683c\u5f0f，\u4e5f\u6ca1\u6709 System Prompt \u548c\u91c7\u6837\u53c2\u6570\u3002\u7b2c\u4e8c\u7ec4\u5305\u542b 20 \u4e2a\u6709\u786e\u5b9a\u7b54\u6848\u7684\u4efb\u52a1，\u7528\u6765\u68c0\u67e5\u66f4\u4f4e\u6210\u672c\u662f\u5426\u4f1a\u4ee5\u727a\u7272\u6b63\u786e\u7387\u4e3a\u4ee3\u4ef7\u3002<\/p>\n

\u6bcf\u4e2a Prompt \u90fd\u4f1a\u9488\u5bf9\u6bcf\u4e2a\u6a21\u578b\u8fd0\u884c 5 \u6b21，\u800c\u4e14\u6267\u884c\u987a\u5e8f\u968f\u673a\u5316\u3002<\/p>\n

\u5728\u770b\u6570\u636e\u4e4b\u524d，\u9700\u8981\u5148\u8bf4\u660e\u4e00\u4e2a\u80cc\u666f\u3002Artificial Analysis \u5728 Intelligence Index \u4e0a\u7ed9 Kimi K3 \u548c GLM-5.3-Flash \u7684\u8bc4\u5206\u90fd\u662f 57\u3002\u4e24\u4e2a\u6a21\u578b\u5728\u540c\u4e00\u4e2a\u72ec\u7acb\u7efc\u5408 Benchmark \u4e0a\u5f97\u5206\u4e00\u6837，\u4f46\u8f93\u5165\u4ef7\u683c\u76f8\u5dee 20 \u500d，\u8f93\u51fa\u4ef7\u683c\u76f8\u5dee 30 \u500d\u3002\u8fd9\u662f\u7b2c\u4e09\u65b9\u6d4b\u91cf\u7ed3\u679c，\u6211\u4eec\u6ca1\u6709\u81ea\u884c\u9a8c\u8bc1\u3002\u4e0d\u8fc7，\u8fd9\u610f\u5473\u7740\u771f\u6b63\u503c\u5f97\u95ee\u7684\u95ee\u9898\u5e76\u4e0d\u662f\u201c\u54ea\u4e2a\u6a21\u578b\u66f4\u806a\u660e\u201d，\u800c\u662f\u201c\u5b83\u4eec\u5404\u81ea\u56de\u7b54\u4e00\u4e2a\u95ee\u9898\u5230\u5e95\u8981\u82b1\u591a\u5c11\u94b1\u201d\u3002<\/p>\n

GLM-5.3-Flash \u5b9e\u9645\u6210\u672c\u53ea\u6bd4 Qwen3.8-Max \u4f4e 2.3 \u500d，\u800c\u4e0d\u662f 12 \u500d<\/h3>\n\n\u6a21\u578b\u5b98\u65b9\u8f93\u51fa\u4ef7\u683c\u5b9e\u6d4b\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u7eb8\u9762\u4ef7\u683c\u6bd4\u5b9e\u6d4b\u4ef7\u683c\u6bd4<\/tr>\n\n\n\n\n
glm-5.3-flash<\/td>\n0.50 \u7f8e\u5143 \/ \u767e\u4e07 Token<\/td>\n0.00156 \u7f8e\u5143<\/td>\n1 \u500d<\/td>\n1 \u500d<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\n6.00 \u7f8e\u5143 \/ \u767e\u4e07 Token<\/td>\n0.00361 \u7f8e\u5143<\/td>\n12 \u500d<\/td>\n2.31 \u500d<\/td>\n<\/tr>\n
kimi-k3<\/td>\n15.00 \u7f8e\u5143 \/ \u767e\u4e07 Token<\/td>\n0.02661 \u7f8e\u5143<\/td>\n30 \u500d<\/td>\n17.01 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Flash \u4f9d\u7136\u662f\u6700\u4fbf\u5b9c\u7684，\u800c\u4e14\u5dee\u8ddd\u5f88\u660e\u663e\u3002\u4f46\u5982\u679c\u4f60\u53ea\u662f\u628a Qwen \u7684\u5355 Token \u4ef7\u683c\u9664\u4ee5 12 \u6765\u4f30\u7b97\u57fa\u7840\u8bbe\u65bd\u9884\u7b97，\u90a3\u4e48\u6700\u7ec8\u7ed3\u679c\u4f1a\u504f\u5dee\u8d85\u8fc7 5 \u500d，\u800c\u4e14\u504f\u5dee\u65b9\u5411\u5bf9\u4f60\u6700\u4e0d\u5229\u3002<\/p>\n

Kimi \u7684\u4ef7\u683c\u4f18\u52bf\u7f29\u6c34\u5f97\u6ca1\u90a3\u4e48\u660e\u663e：\u7eb8\u9762\u4e0a\u662f 30 \u500d，\u5b9e\u9645\u662f 17 \u500d\u3002Qwen \u5219\u7f29\u6c34\u5f97\u66f4\u591a：\u7eb8\u9762 12 \u500d，\u5b9e\u9645\u53ea\u5269 2.3 \u500d\u3002\u8fd9\u79cd\u4e0d\u5bf9\u79f0\u662f\u4e00\u4e2a\u91cd\u8981\u7ebf\u7d22，\u800c\u539f\u56e0\u5c31\u85cf\u5728\u8f93\u51fa\u957f\u5ea6\u91cc\u3002<\/p>\n

\u4e3a\u4ec0\u4e48：GLM-5.3-Flash \u7684\u8f93\u51fa Token \u662f Qwen3.8-Max \u7684 4 \u500d\u4ee5\u4e0a<\/h3>\n

\"images<\/p>\n

\u5728\u5b8c\u5168\u76f8\u540c\u7684 Prompt \u4e0b，\u800c\u4e14\u6ca1\u6709\u7ed9\u4efb\u4f55\u6a21\u578b\u6307\u5b9a\u8f93\u51fa\u683c\u5f0f\u65f6，\u8f93\u51fa Token \u4e2d\u4f4d\u6570\u5982\u4e0b：<\/p>\n\n\u6a21\u578bTier 1（\u7b80\u5355）Tier 2（\u4e2d\u7b49）Tier 3（\u56f0\u96be）<\/tr>\n\n\n\n\n
glm-5.3-flash<\/td>\n625<\/td>\n3141<\/td>\n5018<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\n156<\/td>\n645<\/td>\n886<\/td>\n<\/tr>\n
kimi-k3<\/td>\n456<\/td>\n2006<\/td>\n1961<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5728\u7b80\u5355\u95ee\u9898\u4e0a，Flash \u8f93\u51fa\u7684 Token \u662f Qwen \u7684 4.0 \u500d；\u5230\u4e86\u56f0\u96be\u95ee\u9898，\u8fd9\u4e2a\u6bd4\u4f8b\u8fbe\u5230 5.7 \u500d\u3002\u8fd9\u57fa\u672c\u5df2\u7ecf\u89e3\u91ca\u4e86\u5927\u90e8\u5206\u6210\u672c\u5dee\u8ddd\u3002\u4e00\u4e2a\u6bcf Token \u4fbf\u5b9c 12 \u500d\u3001\u4f46\u8981\u591a\u7528 5 \u500d Token \u7684\u6a21\u578b，\u5b9e\u9645\u5f53\u7136\u4e0d\u53ef\u80fd\u4ecd\u7136\u4fbf\u5b9c 12 \u500d\u3002<\/p>\n

\u8fd9\u4e9b Token \u90fd\u53bb\u4e86\u54ea\u91cc？Flash \u548c Kimi \u90fd\u4f1a\u628a\u63a8\u7406\u8fc7\u7a0b\u5355\u72ec\u653e\u5728 reasoning_content \u5b57\u6bb5\u91cc，\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u76f4\u63a5\u7edf\u8ba1\u3002\u4e0b\u9762\u662f\u8f93\u51fa Token \u4e2d\u7528\u4e8e\u63a8\u7406\u7684\u6bd4\u4f8b\u4e2d\u4f4d\u6570：<\/p>\n\n\u6a21\u578bTier 1Tier 2Tier 3<\/tr>\n\n\n\n\n
glm-5.3-flash<\/td>\n72%<\/td>\n81%<\/td>\n84%<\/td>\n<\/tr>\n
kimi-k3<\/td>\n65%<\/td>\n77%<\/td>\n82%<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\n\u9ed8\u8ba4\u5173\u95ed\u601d\u8003\u6a21\u5f0f<\/td>\n<\/td>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u4e5f\u5c31\u662f\u8bf4，\u8f93\u51fa\u8d26\u5355\u91cc\u5927\u7ea6\u4e09\u5206\u4e4b\u4e8c\u5230\u516d\u5206\u4e4b\u4e94\u7684 Token，\u90fd\u662f\u7528\u6237\u6700\u7ec8\u770b\u4e0d\u5230\u7684\u63a8\u7406\u5185\u5bb9\u3002\u5728 Tier 1 \u548c Tier 2 \u4e2d，Flash \u7684\u63a8\u7406\u5360\u6bd4\u8fd8\u662f\u4e09\u4e2a\u6a21\u578b\u91cc\u6700\u9ad8\u7684，\u800c\u4e14\u968f\u7740\u95ee\u9898\u96be\u5ea6\u63d0\u9ad8，\u8fd9\u4e2a\u6bd4\u4f8b\u8fd8\u4f1a\u7ee7\u7eed\u4e0a\u5347，\u800c\u4e0d\u662f\u4fdd\u6301\u4e0d\u53d8\u3002<\/p>\n

\u5c3e\u90e8\u60c5\u51b5\u6bd4\u4e2d\u4f4d\u6570\u770b\u8d77\u6765\u66f4\u5938\u5f20\u3002Flash \u7684\u5e73\u5747\u8f93\u51fa\u957f\u5ea6\u662f 3111 Token，\u4f46\u4e2d\u4f4d\u6570\u53ea\u6709 1409，\u53d8\u5f02\u7cfb\u6570\u8d85\u8fc7 1.0\u3002\u6211\u4eec\u8bb0\u5f55\u5230\u6700\u957f\u7684\u4e00\u6761 Response \u8fbe\u5230 15,006 Token，\u5176\u4e2d 94.7% \u90fd\u662f\u63a8\u7406 Token，\u5b83\u56de\u7b54\u7684\u662f\u4e00\u4e2a\u90e8\u7f72\u67b6\u6784\u95ee\u9898\u3002\u5355\u72ec\u8fd9\u4e00\u6761 Response \u5c31\u82b1\u4e86 0.00753 \u7f8e\u5143，\u662f\u672c\u6b21\u7814\u7a76\u4e2d Flash \u5e73\u5747\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u7684 4.8 \u500d\u3001\u7b80\u5355\u95ee\u9898\u4e2d\u4f4d\u6210\u672c\u7684 24 \u500d，\u4e5f\u662f reasoning_effort: low \u4e0b\u5e73\u5747\u8bf7\u6c42\u6210\u672c\u7684 30 \u500d\u3002<\/p>\n

\u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48\u672c\u6587\u4e00\u76f4\u4f7f\u7528\u4e2d\u4f4d\u6570，\u800c\u4e0d\u662f\u5e73\u5747\u503c\u3002\u5e73\u5747\u503c\u5728\u8fd9\u91cc\u65e2\u4e0d\u80fd\u8ba9\u6570\u636e\u66f4\u597d\u770b，\u4e5f\u4f1a\u8bef\u5bfc\u8bfb\u8005\u3002<\/p>\n

enable_thinking \u5728 GLM-5.3-Flash\u3001Qwen3.8-Max \u548c Kimi K3 \u4e0a\u7684\u884c\u4e3a\u5b8c\u5168\u4e0d\u540c<\/h3>\n

DigitalOcean \u7684 API \u7aef\u70b9\u517c\u5bb9 OpenAI API，\u5e76\u4e14\u4f1a\u8f6c\u53d1 chat_template_kwargs，\u56e0\u6b64\u4f60\u53ef\u4ee5\u4f20\u5165 enable_thinking: false，\u8ba9\u6a21\u578b\u5c1d\u8bd5\u8df3\u8fc7\u63a8\u7406\u8fc7\u7a0b\u3002\u5f88\u591a\u6587\u7ae0\u90fd\u4f1a\u76f4\u63a5\u628a\u5b83\u5f53\u4f5c\u5173\u95ed\u601d\u8003\u6a21\u5f0f\u7684\u5f00\u5173，\u6211\u4eec\u4e00\u5f00\u59cb\u4e5f\u662f\u8fd9\u4e48\u505a\u7684\u3002<\/p>\n

\u6211\u4eec\u8ba9\u6bcf\u4e2a\u6a21\u578b\u628a\u5b8c\u6574 Prompt Set \u5404\u8dd1\u4e86\u4e09\u904d：\u4e00\u6b21\u5b8c\u5168\u4e0d\u53d1\u9001\u601d\u8003\u53c2\u6570，\u4e00\u6b21\u5c06 enable_thinking \u8bbe\u7f6e\u4e3a\u5f00\u542f，\u4e00\u6b21\u8bbe\u7f6e\u4e3a\u5173\u95ed\u3002\u7ed3\u679c，\u540c\u4e00\u4e2a\u53c2\u6570\u5728\u4e09\u4e2a\u6a21\u578b\u4e0a\u7684\u884c\u4e3a\u5b8c\u5168\u4e0d\u540c\u3002<\/p>\n

\u5728 GLM-5.3-Flash \u4e0a，\u5b83\u57fa\u672c\u53ea\u662f\u6539\u53d8\u4e86\u63a8\u7406\u5185\u5bb9\u5c55\u793a\u7684\u4f4d\u7f6e\u3002 \u628a\u5b83\u8bbe\u4e3a false \u4e4b\u540e，reasoning_content \u5b57\u6bb5\u786e\u5b9e\u53d8\u7a7a\u4e86，\u6240\u4ee5\u65e5\u5fd7\u91cc\u770b\u4e0d\u5230\u63a8\u7406\u8fc7\u7a0b\u4e86\u3002\u4f46\u8f93\u51fa Token \u53cd\u800c\u589e\u52a0\u4e86 20.9%，\u6210\u672c\u4e5f\u589e\u52a0\u4e86 11.1%\u3002<\/p>\n

\u63a8\u7406\u5e76\u6ca1\u6709\u505c\u6b62，\u53ea\u662f\u8dd1\u8fdb\u4e86\u7528\u6237\u80fd\u770b\u5230\u7684\u7b54\u6848\u91cc\u3002\u6bd4\u5982，\u5728\u601d\u8003\u6a21\u5f0f\u7406\u8bba\u4e0a\u5df2\u7ecf\u5173\u95ed\u7684\u60c5\u51b5\u4e0b，\u6a21\u578b\u56de\u7b54\u90e8\u7f72\u62d3\u6251\u95ee\u9898\u65f6，\u5f00\u5934\u662f：<\/p>\n

Let me analyze this problem carefully. The setup: – 70B model – Peak traffic: 40 requests\/second\u2026<\/p>\n

\u518d\u6bd4\u5982，\u56de\u7b54\u201cTTFT \u662f\u4ec0\u4e48\u7f29\u5199\u201d\u8fd9\u79cd\u7b80\u5355\u95ee\u9898\u65f6：<\/p>\n

The user is asking about \u201cTTFT\u201d in the context of LLM inference\u2026 Let me think about what I know about TTFT:<\/p>\n

\u6211\u4eec\u81f3\u5c11\u5728 120 \u6761 Response \u4e2d\u7684 103 \u6761\u91cc\u68c0\u6d4b\u5230\u4e86\u6b8b\u7559\u5728\u53ef\u89c1\u7b54\u6848\u4e2d\u7684 Chain-of-thought\u3002\u8fd9\u4e2a\u68c0\u6d4b\u4f7f\u7528\u7684\u662f\u5173\u952e\u8bcd\u542f\u53d1\u5f0f\u89c4\u5219，\u56e0\u6b64\u5e94\u8be5\u628a 103 \u770b\u4f5c\u4e0b\u9650，\u800c\u4e0d\u662f\u7cbe\u786e\u603b\u6570\u3002\u5728 40 \u6761\u7b80\u5355\u95ee\u9898\u4e2d\u547d\u4e2d\u4e86 38 \u6761，\u5728 40 \u6761\u56f0\u96be\u95ee\u9898\u4e2d\u547d\u4e2d\u4e86 31 \u6761\u3002<\/p>\n

\u5728 Kimi K3 \u4e0a，\u540c\u4e00\u4e2a\u53c2\u6570\u5219\u80fd\u6b63\u5e38\u751f\u6548\u3002 \u8f93\u51fa Token \u4e0b\u964d\u4e86 67.9%，\u6210\u672c\u4e0b\u964d\u4e86 75.5%\u3002\u6bcf\u6b21\u8c03\u7528\u7684\u8f93\u5165\u989d\u5916 Token \u4e5f\u4ece 99 \u4e2a\u964d\u5230 32 \u4e2a，\u56e0\u4e3a Chat Template \u4e2d\u7684\u63a8\u7406\u6a21\u677f\u540c\u6837\u88ab\u79fb\u9664\u4e86\u3002<\/p>\n

\u5728 Qwen3.8-Max \u4e0a，\u601d\u8003\u6a21\u5f0f\u9ed8\u8ba4\u5c31\u662f\u5173\u95ed\u7684\u3002 DigitalOcean \u5f53\u524d\u5c31\u662f\u4ee5\u8fd9\u4e2a\u914d\u7f6e\u63d0\u4f9b Qwen\u3002\u628a\u601d\u8003\u6a21\u5f0f\u6253\u5f00\u540e，\u8f93\u51fa Token \u589e\u52a0\u4e86 429%，\u6210\u672c\u589e\u52a0\u4e86 508%\u3002<\/p>\n

\u6240\u4ee5，enable_thinking \u5e76\u4e0d\u662f\u4e00\u4e2a\u53ef\u4ee5\u8de8\u6a21\u578b\u76f4\u63a5\u590d\u7528\u7684\u7edf\u4e00\u53c2\u6570\u3002\u5728\u4e00\u4e2a\u6a21\u578b\u4e0a，\u5b83\u53ef\u4ee5\u7701 75%；\u5728\u53e6\u4e00\u4e2a\u6a21\u578b\u4e0a，\u5b83\u4f1a\u8ba9\u6210\u672c\u63d0\u9ad8\u7ea6 5 \u500d；\u5230\u4e86\u7b2c\u4e09\u4e2a\u6a21\u578b，\u5b83\u9664\u4e86\u6539\u53d8\u4f60\u80fd\u4e0d\u80fd\u770b\u5230\u63a8\u7406\u8fc7\u7a0b\u4e4b\u5916，\u51e0\u4e4e\u4ec0\u4e48\u4e5f\u6ca1\u6539\u53d8\u3002<\/p>\n

\u800c\u5bf9\u4e8e GLM-5.3-Flash \u6765\u8bf4，\u8fd9\u751a\u81f3\u6839\u672c\u4e0d\u662f\u6b63\u786e\u7684\u63a7\u5236\u53c2\u6570\u3002Model Card \u91cc\u5176\u5b9e\u5199\u4e86\u53e6\u4e00\u4e2a\u53c2\u6570，\u53ea\u662f\u6211\u4eec\u4e00\u5f00\u59cb\u6ca1\u6709\u4ed4\u7ec6\u770b\u5230\u3002<\/p>\n

GLM-5.3-Flash \u7684 reasoning_effort \u9ed8\u8ba4\u662f max，\u8bbe\u7f6e\u6210 low \u540e\u6210\u672c\u964d\u4f4e 84%<\/h3>\n

GLM-5.3-Flash \u901a\u8fc7 reasoning_effort \u63a7\u5236\u63a8\u7406\u5f3a\u5ea6，\u53ef\u8bbe\u7f6e\u4e3a low\u3001high \u6216 max\u3002Model Card \u660e\u786e\u8bf4\u660e：\u5982\u679c\u4e0d\u4f20\u8fd9\u4e2a\u53c2\u6570，\u9ed8\u8ba4\u5c31\u662f max\u3002\u6211\u4eec\u9488\u5bf9\u540c\u6837\u7684 24 \u4e2a Prompt，\u628a\u4e09\u4e2a\u7b49\u7ea7\u90fd\u6d4b\u8bd5\u4e86\u4e00\u904d\u3002<\/p>\n\n\u8bbe\u7f6e\u8f93\u51fa Token \u4e2d\u4f4d\u6570\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u76f8\u6bd4\u9ed8\u8ba4\u914d\u7f6e<\/tr>\n\n\n\n\n\n
\u672a\u8bbe\u7f6e（\u9ed8\u8ba4）<\/td>\n1409<\/td>\n0.001564 \u7f8e\u5143<\/td>\n1.00 \u500d<\/td>\n<\/tr>\n
reasoning_effort: max<\/td>\n1434<\/td>\n0.001526 \u7f8e\u5143<\/td>\n0.98 \u500d<\/td>\n<\/tr>\n
reasoning_effort: high<\/td>\n580<\/td>\n0.000384 \u7f8e\u5143<\/td>\n0.25 \u500d<\/td>\n<\/tr>\n
reasoning_effort: low<\/td>\n414<\/td>\n0.000250 \u7f8e\u5143<\/td>\n0.16 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u8fd9\u91cc\u6709\u4e24\u4e2a\u5173\u952e\u7ed3\u679c\u3002<\/p>\n

\u9ed8\u8ba4\u8bbe\u7f6e\u786e\u5b9e\u5c31\u662f max\u3002 \u5bf9\u540c\u4e00\u4e2a Prompt \u548c\u540c\u4e00\u6b21\u91cd\u590d\u6d4b\u8bd5\u505a\u914d\u5bf9\u6bd4\u8f83\u540e，\u9ed8\u8ba4\u914d\u7f6e\u548c\u663e\u5f0f max \u7684\u4e2d\u4f4d\u6bd4\u503c\u662f 0.984\u3002\u4e5f\u5c31\u662f\u8bf4，\u6bcf\u4e00\u4e2a\u6ca1\u6709\u989d\u5916\u914d\u7f6e\u7684 GLM-5.3-Flash \u8bf7\u6c42，\u90fd\u4f1a\u4f7f\u7528\u5b83\u6700\u6602\u8d35\u7684\u63a8\u7406\u5f3a\u5ea6\u3002<\/p>\n

\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u53ef\u4ee5\u8282\u7701 4～6.3 \u500d\u6210\u672c\u3002 low \u76f8\u6bd4\u9ed8\u8ba4\u914d\u7f6e\u964d\u4f4e\u4e86 84% \u7684\u6210\u672c\u3002\u8fd9\u662f\u6574\u4e2a\u7814\u7a76\u4e2d\u6211\u4eec\u6d4b\u5230\u7684\u6700\u5927\u5355\u9879\u4f18\u5316\u5e45\u5ea6，\u751a\u81f3\u6bd4\u76f4\u63a5\u66f4\u6362\u6a21\u578b\u7684\u5f71\u54cd\u8fd8\u5927\u3002<\/p>\n

\u800c\u4e14\u8fd9\u79cd\u53d8\u5316\u5e76\u4e0d\u662f\u7edf\u4e00\u538b\u7f29，\u800c\u662f\u4f1a\u6839\u636e\u96be\u5ea6\u81ea\u52a8\u8c03\u6574，\u8fd9\u5176\u5b9e\u6b63\u662f\u6211\u4eec\u5e0c\u671b\u770b\u5230\u7684\u884c\u4e3a\u3002\u4e0b\u9762\u662f\u4e0d\u540c\u96be\u5ea6\u4e0b\u63a8\u7406 Token \u5360\u6bd4\u7684\u4e2d\u4f4d\u6570：<\/p>\n\n\u8bbe\u7f6eTier 1（\u7b80\u5355）Tier 2（\u4e2d\u7b49）Tier 3（\u56f0\u96be）<\/tr>\n\n\n\n\n
\u9ed8\u8ba4<\/td>\n72%<\/td>\n81%<\/td>\n84%<\/td>\n<\/tr>\n
high<\/td>\n4%<\/td>\n18%<\/td>\n47%<\/td>\n<\/tr>\n
low<\/td>\n3%<\/td>\n35%<\/td>\n40%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5728 low \u4e0b，\u7b80\u5355\u95ee\u9898\u4e0a\u7684\u63a8\u7406\u57fa\u672c\u6d88\u5931\u4e86\u2014\u2014\u63a8\u7406 Token \u53ea\u5360\u8f93\u51fa\u7684 3%，\u800c\u9ed8\u8ba4\u8bbe\u7f6e\u662f 72%\u3002\u5230\u4e86\u56f0\u96be\u95ee\u9898，\u6a21\u578b\u4ecd\u7136\u4f1a\u8fdb\u884c\u8f83\u591a\u63a8\u7406，\u5927\u7ea6\u5360 40%\u3002\u4e5f\u5c31\u662f\u8bf4，\u8fd9\u5e76\u4e0d\u662f\u628a\u6a21\u578b\u201c\u780d\u6389\u8111\u5b50\u201d，\u800c\u662f\u544a\u8bc9\u5b83\u4e0d\u8981\u5728\u6839\u672c\u4e0d\u9700\u8981\u590d\u6742\u63a8\u7406\u7684\u95ee\u9898\u4e0a\u82b1\u592a\u591a\u65f6\u95f4\u3002<\/p>\n

\u6211\u4eec\u8fd8\u6d4b\u8bd5\u4e86 Model Card \u63a8\u8350\u7528\u4e8e Chat \u573a\u666f\u7684 clear_thinking: true\u3002\u7ed3\u679c\u51e0\u4e4e\u6ca1\u6709\u4efb\u4f55\u53ef\u6d4b\u91cf\u53d8\u5316：\u914d\u5bf9\u6bd4\u503c\u662f 0.99，\u800c\u4e14\u6210\u672c\u8fd8\u7565\u5fae\u66f4\u9ad8\u3002\u8fd9\u4e2a\u7ed3\u679c\u4e5f\u503c\u5f97\u8bb0\u4f4f，\u907f\u514d\u6709\u4eba\u628a\u5b83\u8bef\u5f53\u6210\u4e00\u4e2a\u7701\u94b1\u53c2\u6570\u3002<\/p>\n

\u53ea\u6bd4\u8f83 enable_thinking \u65f6，\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u76f8\u5dee 18.5 \u500d；\u52a0\u5165 reasoning_effort \u540e，\u76f8\u5dee\u8fbe\u5230 106 \u500d<\/h3>\n

\"images<\/p>\n

9 \u79cd\u914d\u7f6e\u2014\u20143 \u4e2a\u6a21\u578b \u00d7 3 \u79cd\u601d\u8003\u6a21\u5f0f\u2014\u2014\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u8de8\u5ea6\u8fbe\u5230 18.5 \u500d\u3002<\/p>\n\n\u6392\u540d\u914d\u7f6e\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u76f8\u6bd4\u6700\u4f4e\u6210\u672c<\/tr>\n\n\n\n\n\n\n\n\n\n\n
1<\/td>\nglm-5.3-flash \/ default<\/td>\n0.00156 \u7f8e\u5143<\/td>\n1.00 \u500d<\/td>\n<\/tr>\n
2<\/td>\nglm-5.3-flash \/ on<\/td>\n0.00157 \u7f8e\u5143<\/td>\n1.01 \u500d<\/td>\n<\/tr>\n
3<\/td>\nglm-5.3-flash \/ off<\/td>\n0.00174 \u7f8e\u5143<\/td>\n1.11 \u500d<\/td>\n<\/tr>\n
4<\/td>\nqwen3.8-max \/ off<\/td>\n0.00359 \u7f8e\u5143<\/td>\n2.29 \u500d<\/td>\n<\/tr>\n
5<\/td>\nqwen3.8-max \/ default<\/td>\n0.00361 \u7f8e\u5143<\/td>\n2.31 \u500d<\/td>\n<\/tr>\n
6<\/td>\nkimi-k3 \/ off<\/td>\n0.00651 \u7f8e\u5143<\/td>\n4.17 \u500d<\/td>\n<\/tr>\n
7<\/td>\nqwen3.8-max \/ on<\/td>\n0.02196 \u7f8e\u5143<\/td>\n14.04 \u500d<\/td>\n<\/tr>\n
8<\/td>\nkimi-k3 \/ default<\/td>\n0.02661 \u7f8e\u5143<\/td>\n17.01 \u500d<\/td>\n<\/tr>\n
9<\/td>\nkimi-k3 \/ on<\/td>\n0.02892 \u7f8e\u5143<\/td>\n18.49 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u628a\u7b2c 6 \u884c\u548c\u7b2c 7 \u884c\u653e\u5728\u4e00\u8d77\u770b\u3002\u914d\u7f6e\u5f97\u5f53\u7684 Kimi K3，\u5b9e\u9645\u6210\u672c\u751a\u81f3\u4f4e\u4e8e\u914d\u7f6e\u9519\u8bef\u7684 Qwen3.8-Max，\u5c3d\u7ba1 Kimi \u7684\u5b98\u65b9\u8f93\u51fa\u4ef7\u683c\u662f Qwen \u7684 2.5 \u500d\u3002\u4ec5\u4ec5\u4e00\u4e2a\u5e03\u5c14\u53c2\u6570，\u5c31\u80fd\u8ba9 Kimi \u4ece 17.01 \u500d\u964d\u5230 4.17 \u500d\u2014\u2014\u8fd9\u4e2a\u53d8\u5316\u5e45\u5ea6\u751a\u81f3\u6bd4\u76f4\u63a5\u6362\u6a21\u578b\u8fd8\u5927\u3002<\/p>\n

Flash \u7684\u4e09\u79cd enable_thinking \u914d\u7f6e\u4e4b\u95f4\u6210\u672c\u5dee\u8ddd\u53ea\u6709 11%\u3002\u8fd9\u5e76\u4e0d\u662f\u56e0\u4e3a Flash \u6ca1\u6709\u53ef\u8c03\u8282\u7684\u53c2\u6570，\u800c\u662f\u56e0\u4e3a\u6211\u4eec\u4e00\u5f00\u59cb\u8c03\u9519\u4e86\u53c2\u6570\u3002\u628a reasoning_effort \u52a0\u8fdb\u6765\u4e4b\u540e，\u540c\u4e00\u5f20\u6210\u672c\u9636\u68af\u5c31\u53d8\u6210\u4e86\u8fd9\u6837：<\/p>\n\n\u914d\u7f6e\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u76f8\u6bd4\u6700\u4f4e\u6210\u672c<\/tr>\n\n\n\n\n\n\n\n
glm-5.3-flash \/ reasoning_effort: low<\/td>\n0.000250 \u7f8e\u5143<\/td>\n1.00 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash \/ reasoning_effort: high<\/td>\n0.000384 \u7f8e\u5143<\/td>\n1.54 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash \/ default<\/td>\n0.001564 \u7f8e\u5143<\/td>\n6.26 \u500d<\/td>\n<\/tr>\n
qwen3.8-max \/ default<\/td>\n0.003610 \u7f8e\u5143<\/td>\n14.44 \u500d<\/td>\n<\/tr>\n
kimi-k3 \/ off<\/td>\n0.006510 \u7f8e\u5143<\/td>\n26.04 \u500d<\/td>\n<\/tr>\n
kimi-k3 \/ default<\/td>\n0.026610 \u7f8e\u5143<\/td>\n106.44 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u6b64\u65f6\u6210\u672c\u8de8\u5ea6\u4e0d\u518d\u662f 18.5 \u500d，\u800c\u662f 106 \u500d\u3002\u5176\u4e2d\u6700\u5927\u7684\u5355\u4e00\u5f71\u54cd\u56e0\u7d20，\u53ea\u662f\u4e00\u4e2a\u6a21\u578b\u4e0a\u7684\u4e00\u4e2a\u53c2\u6570\u3002<\/p>\n

\u8fd9\u91cc\u8fd8\u6709\u4e00\u4e2a\u503c\u5f97\u5355\u72ec\u5f3a\u8c03\u7684\u4e8c\u9636\u5f71\u54cd：\u4e0d\u540c\u6a21\u578b\u7684\u9ed8\u8ba4\u914d\u7f6e\u6839\u672c\u4e0d\u4e00\u81f4\u2014\u2014Qwen \u9ed8\u8ba4\u5173\u95ed\u601d\u8003\u6a21\u5f0f，\u800c Flash \u9ed8\u8ba4\u4f7f\u7528\u6700\u9ad8\u63a8\u7406\u5f3a\u5ea6\u3002\u56e0\u6b64，\u5982\u679c\u4f60\u62ff\u51e0\u4e2a\u6a21\u578b\u76f4\u63a5\u201c\u5f00\u7bb1\u5373\u7528\u201d\u505a\u5bf9\u6bd4，\u5b9e\u9645\u4e0a\u6bd4\u8f83\u7684\u5e76\u4e0d\u53ea\u662f\u6a21\u578b，\u800c\u662f\u6a21\u578b + \u9ed8\u8ba4\u914d\u7f6e\u3002\u7edd\u5927\u591a\u6570\u516c\u5f00\u6a21\u578b\u5bf9\u6bd4\u90fd\u4f1a\u8fd9\u4e48\u505a，\u5305\u62ec\u6211\u4eec\u81ea\u5df1\u5728\u8865\u6d4b\u8fd9\u4e9b\u914d\u7f6e\u4e4b\u524d\u4e5f\u662f\u5982\u6b64\u3002<\/p>\n

\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u4f1a\u5f71\u54cd\u51c6\u786e\u7387\u5417？\u4f1a，\u4f46\u53ea\u51fa\u73b0\u5728\u56f0\u96be\u95ee\u9898\u4e0a<\/h3>\n

\"images<\/p>\n

\u770b\u5230\u8fd9\u91cc，\u4e00\u4e2a\u5f88\u81ea\u7136\u7684\u8d28\u7591\u662f：\u4e5f\u8bb8\u8d35\u6a21\u578b\u4e4b\u6240\u4ee5\u8d35，\u5c31\u662f\u56e0\u4e3a\u80fd\u529b\u66f4\u5f3a，\u800c\u6211\u4eec\u6d4b\u5230\u7684\u53ea\u662f\u201c\u66f4\u9ad8\u8d28\u91cf\u7b54\u6848\u6240\u5bf9\u5e94\u7684\u4ef7\u683c\u201d\u3002<\/p>\n

\u6240\u4ee5\u6211\u4eec\u53c8\u6784\u5efa\u4e86\u7b2c\u4e8c\u5957\u4efb\u52a1：20 \u4e2a\u6709\u786e\u5b9a\u7b54\u6848\u7684\u95ee\u9898，\u5305\u62ec\u6210\u672c\u8ba1\u7b97\u3001Retry Policy \u4e0b\u7684\u671f\u671b\u503c\u3001\u7ec4\u5408\u6570\u5b66\u95ee\u9898，\u4ee5\u53ca\u5ef6\u8fdf\u9884\u7b97\u63a8\u5bfc\u3002\u6bcf\u4e2a\u95ee\u9898\u90fd\u8981\u6c42\u6700\u540e\u4e00\u884c\u4e25\u683c\u4f7f\u7528 ANSWER: \u56fa\u5b9a\u683c\u5f0f，\u8fd9\u6837\u8bc4\u5206\u53ef\u4ee5\u76f4\u63a5\u901a\u8fc7\u5b57\u7b26\u4e32\u63d0\u53d6\u5b8c\u6210，\u800c\u4e0d\u4f9d\u8d56\u4eba\u5de5\u5224\u65ad\u3002\u6d4b\u8bd5\u8986\u76d6 3 \u4e2a\u6a21\u578b\u30013 \u79cd\u601d\u8003\u6a21\u5f0f\u548c 2 \u79cd\u964d\u4f4e\u540e\u7684\u63a8\u7406\u5f3a\u5ea6，\u6bcf\u4e2a\u914d\u7f6e\u8fd0\u884c 5 \u6b21，\u603b\u5171\u5f97\u5230 1099 \u6761\u53ef\u8bc4\u5206 Response\u3002<\/p>\n

\u5728\u9ed8\u8ba4\u914d\u7f6e\u4e0b，\u4e09\u4e2a\u6a21\u578b\u5168\u90e8\u62ff\u5230\u4e86 100%\u3002<\/p>\n\n\u6a21\u578b\u8bbe\u7f6epass@1\u5355\u4e2a\u6b63\u786e\u7b54\u6848\u6210\u672c\u76f8\u6bd4\u6700\u4f4e\u6210\u672c<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n\n
glm-5.3-flash<\/td>\neffort: low<\/td>\n92.0%<\/td>\n0.00008 \u7f8e\u5143<\/td>\n1.00 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash<\/td>\neffort: high<\/td>\n98.0%<\/td>\n0.00012 \u7f8e\u5143<\/td>\n1.47 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash<\/td>\nenable_thinking: false<\/td>\n100.0%<\/td>\n0.00018 \u7f8e\u5143<\/td>\n2.15 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash<\/td>\ndefault<\/td>\n100.0%<\/td>\n0.00018 \u7f8e\u5143<\/td>\n2.17 \u500d<\/td>\n<\/tr>\n
glm-5.3-flash<\/td>\nenable_thinking: true<\/td>\n100.0%<\/td>\n0.00019 \u7f8e\u5143<\/td>\n2.26 \u500d<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\nenable_thinking: false<\/td>\n100.0%<\/td>\n0.00226 \u7f8e\u5143<\/td>\n26.64 \u500d<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\ndefault<\/td>\n100.0%<\/td>\n0.00235 \u7f8e\u5143<\/td>\n27.65 \u500d<\/td>\n<\/tr>\n
kimi-k3<\/td>\nenable_thinking: false<\/td>\n98.0%<\/td>\n0.00282 \u7f8e\u5143<\/td>\n33.23 \u500d<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\nenable_thinking: true<\/td>\n100.0%<\/td>\n0.00383 \u7f8e\u5143<\/td>\n45.14 \u500d<\/td>\n<\/tr>\n
kimi-k3<\/td>\ndefault<\/td>\n100.0%<\/td>\n0.00565 \u7f8e\u5143<\/td>\n66.47 \u500d<\/td>\n<\/tr>\n
kimi-k3<\/td>\nenable_thinking: true<\/td>\n100.0%<\/td>\n0.00601 \u7f8e\u5143<\/td>\n70.78 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5728 9 \u79cd enable_thinking \u914d\u7f6e\u4e2d，\u5404\u6a21\u578b\u7684\u51c6\u786e\u7387\u90fd\u8fbe\u5230\u6216\u63a5\u8fd1 100%，\u4f46\u5355\u4e2a\u6b63\u786e\u7b54\u6848\u6210\u672c\u76f8\u5dee\u4e86 33 \u500d\u3002\u5728\u8fd9\u4e2a\u6d4b\u8bd5\u8303\u56f4\u5185，\u4fbf\u5b9c\u6a21\u578b\u5e76\u4e0d\u662f\u56e0\u4e3a\u80fd\u529b\u5dee\u624d\u4fbf\u5b9c\u3002\u81f3\u5c11\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e0a，\u5b83\u5e76\u6ca1\u6709\u66f4\u5dee\u3002<\/p>\n

\u4f46\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u5e76\u4e0d\u662f\u6ca1\u6709\u4ee3\u4ef7\u3002 Flash \u5728\u9ed8\u8ba4\u8bbe\u7f6e\u4e0b\u5f97\u5206 100%，high \u662f 98%，\u800c low \u964d\u5230 92%\u3002\u4e5f\u5c31\u662f\u8bf4，\u7701\u94b1\u786e\u5b9e\u4f1a\u5e26\u6765\u4e00\u5b9a\u51c6\u786e\u7387\u635f\u5931，\u800c\u4e14\u6211\u4eec\u80fd\u660e\u786e\u770b\u5230\u635f\u5931\u53d1\u751f\u5728\u54ea\u91cc：\u6240\u6709\u9519\u8bef\u90fd\u51fa\u73b0\u5728 Tier 3 \u4efb\u52a1\u4e2d\u3002Tier 1 \u548c Tier 2 \u5728\u6240\u6709\u63a8\u7406\u5f3a\u5ea6\u4e0b\u90fd\u662f\u6ee1\u5206\u3002<\/p>\n\n\u8bbe\u7f6e\u5931\u8d25\u6b21\u6570\u5bf9\u5e94\u4efb\u52a1<\/tr>\n\n\n\n\n
\u9ed8\u8ba4<\/td>\n0\/100<\/td>\n\u2014<\/td>\n<\/tr>\n
high<\/td>\n2\/100<\/td>\n\u6982\u7387（1）\u3001\u903b\u8f91（1）<\/td>\n<\/tr>\n
low<\/td>\n8\/100<\/td>\n\u6210\u672c\u5efa\u6a21（3）\u3001\u6982\u7387（2）\u3001\u903b\u8f91（3）<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5728 low \u4e0b，\u4e00\u4e2a\u591a\u6b65\u6210\u672c\u5efa\u6a21\u95ee\u9898 5 \u6b21\u91cc\u9519\u4e86 3 \u6b21\u2014\u2014\u5176\u4e2d\u4e24\u6b21\u628a\u6b63\u786e\u7b54\u6848 61.22 \u5199\u6210\u4e86 0.6122，\u8fd9\u66f4\u50cf\u662f\u5355\u4f4d\u9519\u8bef，\u800c\u4e0d\u662f\u7eaf\u7cb9\u7684\u63a8\u7406\u9519\u8bef\u3002\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u5e76\u4e0d\u4f1a\u8ba9\u6a21\u578b\u7a81\u7136\u4e0d\u4f1a\u8bb0\u5fc6\u6216\u7b97\u672f，\u800c\u662f\u4f1a\u964d\u4f4e\u5b83\u7ef4\u6301\u591a\u6b65\u63a8\u7406\u94fe\u6761\u3001\u5e76\u5728\u6700\u540e\u68c0\u67e5\u7ed3\u679c\u7684\u80fd\u529b\u3002<\/p>\n

\u4ece\u5355\u4e2a\u6b63\u786e\u7b54\u6848\u6210\u672c\u6765\u770b，low \u4f9d\u7136\u66f4\u5212\u7b97，\u56e0\u4e3a\u5728\u8fd9\u5957\u76f8\u5bf9\u7b80\u5355\u7684\u6d4b\u8bd5\u91cc，8% \u7684\u5931\u8d25\u7387\u4ecd\u7136\u62b5\u4e0d\u8fc7 84% \u7684\u6210\u672c\u4e0b\u964d\u3002\u4f46\u6211\u4eec\u4e0d\u4f1a\u628a\u8fd9\u4e2a\u7ed3\u8bba\u76f4\u63a5\u63a8\u5e7f\u5230\u66f4\u590d\u6742\u7684\u4efb\u52a1\u4e0a\u3002\u8fd9\u91cc\u6700\u91cd\u8981\u7684\u9650\u5236\u662f：\u6211\u4eec\u7684\u4efb\u52a1\u96c6\u5bf9\u6a21\u578b\u6765\u8bf4\u8fd8\u662f\u592a\u7b80\u5355，11 \u79cd\u914d\u7f6e\u91cc\u6709 6 \u79cd\u8fbe\u5230 100%\u3002\u8fd9\u6837\u7684\u6d4b\u8bd5\u8db3\u4ee5\u8bc1\u660e\u201c\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u4f1a\u5e26\u6765\u51c6\u786e\u7387\u4e0b\u964d\u201d，\u4f46\u4e0d\u8db3\u4ee5\u544a\u8bc9\u4f60\u5728\u771f\u6b63\u56f0\u96be\u7684\u4efb\u52a1\u4e0a\u4f1a\u4e0b\u964d\u591a\u5c11\u3002\u5bf9\u4e8e\u66f4\u96be\u7684\u63a8\u7406\u4efb\u52a1，\u53ef\u4ee5\u9884\u671f\u5dee\u8ddd\u4f1a\u8fdb\u4e00\u6b65\u6269\u5927\u3002<\/p>\n

\u7ed3\u6784\u5316\u8f93\u51fa\u53ef\u4ee5\u8ba9 GLM-5.3-Flash \u91cd\u65b0\u83b7\u5f97\u63a5\u8fd1\u7eb8\u9762\u4ef7\u683c\u7684\u6210\u672c\u4f18\u52bf<\/h3>\n

\u4e24\u7ec4 Prompt \u4e4b\u95f4\u6709\u4e00\u4e2a\u91cd\u8981\u533a\u522b：\u53ef\u9a8c\u8bc1\u4efb\u52a1\u5f3a\u5236\u8981\u6c42\u56fa\u5b9a\u8f93\u51fa\u683c\u5f0f，\u800c\u5f00\u653e\u5f0f Prompt \u6ca1\u6709\u4efb\u4f55\u683c\u5f0f\u9650\u5236\u3002\u8fd9\u4e2a\u539f\u672c\u5e76\u4e0d\u662f\u4e3b\u8981\u6d4b\u8bd5\u53d8\u91cf\u7684\u5dee\u5f02，\u6700\u540e\u5374\u53d8\u6210\u4e86\u6574\u4e2a\u7814\u7a76\u4e2d\u6700\u6709\u4ef7\u503c\u7684\u7ed3\u679c\u4e4b\u4e00\u3002<\/p>\n\n\u6a21\u578b\u5f00\u653e\u5f0f\u8f93\u51fa Token \u4e2d\u4f4d\u6570\u7ed3\u6784\u5316\u8f93\u51fa Token \u4e2d\u4f4d\u6570\u964d\u5e45<\/tr>\n\n\n\n\n
glm-5.3-flash<\/td>\n1409<\/td>\n276<\/td>\n80.4%<\/td>\n<\/tr>\n
kimi-k3<\/td>\n1133<\/td>\n248<\/td>\n78.1%<\/td>\n<\/tr>\n
qwen3.8-max<\/td>\n435<\/td>\n144<\/td>\n66.9%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u7ed3\u6784\u5316\u8f93\u51fa\u5bf9 Flash \u7684\u538b\u7f29\u6548\u679c\u6bd4\u53e6\u5916\u4e24\u4e2a\u6a21\u578b\u66f4\u660e\u663e，\u800c Flash \u7684\u5197\u957f\u8f93\u51fa\u6070\u6070\u5c31\u662f\u6b64\u524d\u4fb5\u8680\u5b83\u4ef7\u683c\u4f18\u52bf\u7684\u4e3b\u8981\u539f\u56e0\u3002\u56e0\u6b64，Flash \u7684\u5b9e\u9645\u6210\u672c\u4f18\u52bf\u4f1a\u968f\u7740\u5de5\u4f5c\u8d1f\u8f7d\u5f62\u5f0f\u53d8\u5316：<\/p>\n\nGLM-5.3-Flash \u76f8\u6bd4\u5f00\u653e\u5f0f Prompt\u7ed3\u6784\u5316\u8f93\u51fa<\/tr>\n\n\n\n
qwen3.8-max<\/td>\n2.3 \u500d<\/td>\n12.8 \u500d<\/td>\n<\/tr>\n
kimi-k3<\/td>\n17.0 \u500d<\/td>\n30.7 \u500d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5728\u7ed3\u6784\u5316\u8f93\u51fa\u4e0b，Flash \u51e0\u4e4e\u5b8c\u5168\u5151\u73b0\u4e86\u5b98\u65b9\u4ef7\u683c\u6697\u793a\u7684\u6210\u672c\u4f18\u52bf；\u800c\u5728\u5b8c\u5168\u4e0d\u53d7\u7ea6\u675f\u7684\u5f00\u653e\u5f0f\u8f93\u51fa\u4e2d，\u5b83\u76f8\u5bf9 Qwen \u53ea\u80fd\u5151\u73b0\u5927\u7ea6\u4e94\u5206\u4e4b\u4e00\u3002<\/p>\n

\u4f46\u8fd9\u91cc\u4e5f\u6709\u4e00\u4e2a\u4e0d\u80fd\u5ffd\u7565\u7684\u9650\u5236\u3002\u6211\u4eec\u8fd8\u628a 8 \u4e2a\u7b80\u5355\u95ee\u9898\u91cd\u65b0\u6d4b\u8bd5\u4e86\u4e00\u904d，\u53ea\u5728\u7ed3\u5c3e\u52a0\u4e0a\u4e00\u53e5 \u201cAnswer in one sentence.\u201d\u3002\u7ed3\u679c\u4e09\u4e2a\u6a21\u578b\u7684\u8f93\u51fa Token \u90fd\u4e0b\u964d\u4e86 70%～91%，\u6548\u679c\u751a\u81f3\u6bd4\u601d\u8003\u53c2\u6570\u5728\u5176\u4e2d\u4e24\u4e2a\u6a21\u578b\u4e0a\u7684\u5f71\u54cd\u8fd8\u5927\u3002\u4f46\u63a8\u7406 Token \u5360\u6bd4\u51e0\u4e4e\u6ca1\u53d8\u5316：Flash \u4ece 72% \u53d8\u6210 77%，Kimi \u5219\u59cb\u7ec8\u662f 65%\u3002<\/p>\n

\u4e5f\u5c31\u662f\u8bf4，\u8fd9\u6761\u6307\u4ee4\u538b\u7f29\u7684\u662f\u6700\u7ec8\u7b54\u6848，\u800c\u4e0d\u662f\u6a21\u578b\u5185\u90e8\u7684\u63a8\u7406\u8fc7\u7a0b\u3002\u4f60\u4ecd\u7136\u4f1a\u628a\u76f8\u540c\u6bd4\u4f8b\u7684\u8d26\u5355\u82b1\u5728\u7528\u6237\u770b\u4e0d\u5230\u7684\u63a8\u7406\u4e0a，\u53ea\u662f\u6574\u4e2a\u8d26\u5355\u672c\u8eab\u53d8\u5c0f\u4e86\u3002<\/p>\n

\u4e0d\u540c\u5de5\u4f5c\u8d1f\u8f7d\u5e94\u8be5\u4f7f\u7528\u4ec0\u4e48 reasoning_effort？<\/h3>\n

\u5728 GLM-5.3-Flash \u4e0a\u663e\u5f0f\u8bbe\u7f6e reasoning_effort\u3002 \u5982\u679c\u4e0d\u8bbe\u7f6e，\u5b83\u5c31\u4f1a\u4f7f\u7528 max，\u4e5f\u5c31\u662f\u6700\u6602\u8d35\u7684\u914d\u7f6e\u3002\u8fd9\u662f\u6574\u4e2a\u6d4b\u8bd5\u4e2d\u5f71\u54cd\u6700\u5927\u7684\u5355\u4e00\u6210\u672c\u8c03\u8282\u624b\u6bb5\u3002<\/p>\n

\u5bf9\u4e8e\u5206\u7c7b\u3001\u4fe1\u606f\u63d0\u53d6\u3001\u6a21\u578b\u8def\u7531\u3001\u683c\u5f0f\u8f6c\u6362\u548c\u7b80\u5355\u67e5\u8be2，\u4f7f\u7528 reasoning_effort: low\u3002 \u5728\u6240\u6709\u4f4e\u4e8e\u590d\u6742\u591a\u6b65\u63a8\u7406\u7684\u4efb\u52a1\u4e0a，\u5b83\u6ca1\u6709\u5e26\u6765\u51c6\u786e\u7387\u635f\u5931，\u4f46\u76f8\u6bd4\u9ed8\u8ba4\u914d\u7f6e\u4fbf\u5b9c\u4e86 6.3 \u500d\u3002<\/p>\n

\u5bf9\u4e8e\u6df7\u5408\u5de5\u4f5c\u8d1f\u8f7d，high \u6bd4\u9ed8\u8ba4\u914d\u7f6e\u66f4\u9002\u5408\u4f5c\u4e3a\u9ed8\u8ba4\u503c\u3002 \u5728\u6211\u4eec\u7684\u4efb\u52a1\u96c6\u4e0a，\u5b83\u76f8\u6bd4 max \u4fbf\u5b9c 4 \u500d，\u51c6\u786e\u7387\u662f 98%，\u800c\u9ed8\u8ba4\u914d\u7f6e\u662f 100%\u3002\u5982\u679c\u4f60\u53ea\u60f3\u9009\u62e9\u4e00\u4e2a\u8bbe\u7f6e\u957f\u671f\u4f7f\u7528，\u4f18\u5148\u8003\u8651\u8fd9\u4e2a\u3002<\/p>\n

\u5bf9\u4e8e\u771f\u6b63\u56f0\u96be\u7684\u591a\u6b65\u63a8\u7406，\u4fdd\u6301 max\u3002 \u6211\u4eec\u5728\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u540e\u770b\u5230\u7684\u6240\u6709\u9519\u8bef\u90fd\u51fa\u73b0\u5728 Tier 3，\u800c\u6211\u4eec\u7684\u6d4b\u8bd5\u96c6\u672c\u8eab\u8fd8\u4e0d\u591f\u96be，\u65e0\u6cd5\u51c6\u786e\u544a\u8bc9\u4f60\u5728\u66f4\u590d\u6742\u4efb\u52a1\u4e2d\u6027\u80fd\u4f1a\u4e0b\u964d\u591a\u5c11\u3002<\/p>\n

\u5982\u679c\u8f93\u51fa\u662f\u5f00\u653e\u5f0f\u4e14\u6ca1\u6709\u957f\u5ea6\u9650\u5236，\u9884\u7b97\u5e94\u8be5\u57fa\u4e8e\u5b9e\u9645 Token，\u800c\u4e0d\u662f\u5b98\u65b9\u4ef7\u683c\u3002 \u5148\u8dd1 100 \u6b21\u5177\u6709\u4ee3\u8868\u6027\u7684\u771f\u5b9e\u8bf7\u6c42，\u518d\u6839\u636e\u5b9e\u9645 Token \u91cf\u4f30\u7b97\u3002<\/p>\n

\u5982\u679c\u4f60\u5df2\u7ecf\u5728\u4f7f\u7528 Kimi K3，\u53ef\u4ee5\u6d4b\u8bd5\u5de5\u4f5c\u8d1f\u8f7d\u662f\u5426\u80fd\u63a5\u53d7 enable_thinking: false\u3002 \u5728\u6211\u4eec\u7684\u6d4b\u8bd5\u91cc，\u5b83\u8282\u7701\u4e86 75% \u6210\u672c，\u4f46\u51c6\u786e\u7387\u4e0b\u964d\u4e86 2 \u4e2a\u767e\u5206\u70b9\u3002<\/p>\n

\u5982\u679c\u4f60\u6b63\u5728\u4f7f\u7528 Qwen3.8-Max，\u5e76\u8003\u8651\u6253\u5f00\u601d\u8003\u6a21\u5f0f，\u8981\u5148\u6ce8\u610f\u5b83\u4f1a\u8ba9\u6210\u672c\u589e\u52a0\u7ea6 5 \u500d\u3002 \u5f00\u542f\u4e4b\u524d，\u6700\u597d\u5148\u9a8c\u8bc1\u5b83\u662f\u5426\u771f\u7684\u80fd\u5e26\u6765\u5b9e\u9645\u6536\u76ca\u3002<\/p>\n

\u4e0d\u8981\u5728 GLM-5.3-Flash \u4e0a\u4f7f\u7528 enable_thinking: false \u5e76\u671f\u5f85\u5b83\u80fd\u7701\u94b1\u3002\u6211\u4eec\u7684\u6d4b\u8bd5\u4e2d，\u5b83\u53cd\u800c\u8ba9\u6210\u672c\u589e\u52a0\u4e86 11%\u3002\u540c\u6837，\u4e5f\u4e0d\u8981\u6307\u671b clear_thinking: true \u80fd\u6539\u53d8\u6210\u672c，\u56e0\u4e3a\u5b9e\u6d4b\u6ca1\u6709\u6548\u679c\u3002<\/p>\n

\u76ee\u524d，\u8fd9\u4e09\u4e2a\u6a21\u578b\u90fd\u5df2\u7ecf\u53ef\u4ee5\u901a\u8fc7 DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406 \u4f7f\u7528\u3002API \u7aef\u70b9\u517c\u5bb9 OpenAI，\u5730\u5740\u662f https:\/\/inference.do-ai.run\/v1，\u56e0\u6b64\u8981\u5728\u81ea\u5df1\u7684\u5de5\u4f5c\u8d1f\u8f7d\u4e0a\u590d\u73b0\u8fd9\u4e9b\u6d4b\u8bd5，\u57fa\u672c\u53ea\u9700\u8981\u4fee\u6539 Base URL \u548c\u6a21\u578b ID：glm-5.3-flash\u3001qwen3.8-max \u6216 kimi-k3\u3002\u672c\u6587\u6240\u6709\u6d4b\u8bd5\u90fd\u662f\u901a\u8fc7\u8fd9\u4e2a API \u7aef\u70b9，\u4f7f\u7528\u6807\u51c6 openai Python Client \u5b8c\u6210\u7684\u3002<\/p>\n

\u4e3a\u4ec0\u4e48\u8981\u5728 DigitalOcean \u4e0a\u8fd0\u884c GLM-5.3-Flash，\u800c\u4e0d\u662f\u5176\u4ed6\u5e73\u53f0？<\/h3>\n

\u5148\u5766\u8bda\u8bf4\u660e\u4e00\u70b9：\u672c\u6587\u4f7f\u7528\u7684\u4ef7\u683c\u5e76\u4e0d\u662f DigitalOcean \u72ec\u6709\u3002GLM-5.3-Flash \u91c7\u7528 MIT \u8bb8\u53ef\u8bc1，Z.ai \u516c\u5e03\u7684\u5b98\u65b9\u4ef7\u683c\u4e0e\u6211\u4eec\u6d4b\u8bd5\u4f7f\u7528\u7684\u4ef7\u683c\u4e00\u81f4，\u56e0\u6b64\u5355\u7eaf\u770b\u6a21\u578b\u7684\u5355 Token \u6210\u672c，\u5e76\u4e0d\u8db3\u4ee5\u6210\u4e3a\u9009\u62e9\u67d0\u4e2a\u6258\u7ba1\u5e73\u53f0\u7684\u7406\u7531\u3002\u5982\u679c\u6709\u4eba\u4ee5\u660e\u663e\u66f4\u4f4e\u7684\u4ef7\u683c\u63d0\u4f9b\u5b8c\u5168\u76f8\u540c\u7684\u6a21\u578b\u6743\u91cd，\u8981\u4e48\u662f\u5e73\u53f0\u81ea\u5df1\u505a\u4e86\u6298\u6263，\u8981\u4e48\u6bd4\u8f83\u7684\u5176\u5b9e\u4e0d\u662f\u540c\u4e00\u4ef6\u4e1c\u897f\u3002<\/p>\n

\u771f\u6b63\u4e0d\u540c\u7684，\u662f Token \u5468\u56f4\u7684\u6574\u5957\u57fa\u7840\u8bbe\u65bd\u3002\u8fd9\u6b21\u6d4b\u8bd5\u8fc7\u7a0b\u4e2d，\u6709\u4e09\u4e2a\u56e0\u7d20\u5f88\u91cd\u8981，\u800c\u5b83\u4eec\u5728\u751f\u4ea7\u73af\u5883\u91cc\u540c\u6837\u91cd\u8981\u3002<\/p>\n

\u5468\u8fb9\u6280\u672f\u6808\u90fd\u5728\u540c\u4e00\u4e2a\u5730\u65b9\u3002 Token \u4ef7\u683c\u53ea\u662f\u63a8\u7406\u8d26\u5355\u7684\u4e00\u90e8\u5206\u3002\u5982\u679c\u5411\u91cf\u6570\u636e\u5e93\u3001\u5bf9\u8c61\u5b58\u50a8\u548c\u5e94\u7528\u670d\u52a1\u5668\u5728\u4e00\u5bb6 Cloud，\u800c\u6a21\u578b\u90e8\u7f72\u5728\u53e6\u4e00\u5bb6，\u90a3\u4e48\u6bcf\u6b21 Retrieval \u90fd\u9700\u8981\u652f\u4ed8\u51fa\u7ad9\u6d41\u91cf\u8d39\u7528，\u8fd8\u8981\u627f\u62c5\u8de8\u5e73\u53f0\u5f80\u8fd4\u5ef6\u8fdf\u3002\u5bf9\u4e8e\u4e00\u4e2a\u6bcf\u6b21\u8bf7\u6c42\u9700\u8981\u8fdb\u884c\u591a\u6b21 Retrieval \u7684 RAG Pipeline \u6765\u8bf4，\u8fd9\u79cd\u8de8\u5e73\u53f0\u8bf7\u6c42\u589e\u52a0\u7684\u5ef6\u8fdf，\u751a\u81f3\u53ef\u80fd\u9ad8\u4e8e\u6a21\u578b\u81ea\u8eab\u7684\u9996 Token \u5ef6\u8fdf（TTFT）\u2014\u2014\u800c\u8fd9\u4e9b\u6210\u672c\u548c\u5ef6\u8fdf\u90fd\u4e0d\u4f1a\u51fa\u73b0\u5728\u5355 Token \u4ef7\u683c\u6bd4\u8f83\u91cc\u3002\u5982\u679c\u63a8\u7406\u670d\u52a1\u548c Managed Postgres\u3001Spaces\u3001App Platform \u90e8\u7f72\u5728\u540c\u4e00\u5e73\u53f0\u4e0a，\u5c31\u53ef\u4ee5\u907f\u514d\u8fd9\u4e00\u6b21\u8de8\u5e73\u53f0\u8df3\u8f6c\u3002<\/p>\n

\u4e00\u4e2a\u8d26\u53f7，\u4e00\u5f20\u8d26\u5355\u3002 \u65e0\u670d\u52a1\u5668\u63a8\u7406\u7684\u4f7f\u7528\u91cf\u548c\u5176\u4ed6 DigitalOcean \u57fa\u7840\u8bbe\u65bd\u90fd\u4f1a\u51fa\u73b0\u5728\u540c\u4e00\u4e2a DigitalOcean \u8d26\u53f7\u548c\u540c\u4e00\u5f20\u8d26\u5355\u4e0a\u3002\u8fd9\u542c\u8d77\u6765\u53ea\u662f\u7ba1\u7406\u4e0a\u7684\u4fbf\u5229，\u4f46\u5982\u679c\u4f60\u771f\u6b63\u5c1d\u8bd5\u8fc7\u5ba1\u8ba1\u4e00\u5f20\u62c6\u5728\u4e09\u4e2a\u5e73\u53f0\u3001\u4f7f\u7528\u4e09\u79cd\u8ba1\u8d39\u6a21\u5f0f\u548c\u4e09\u5957 Token \u7edf\u8ba1\u89c4\u5219\u7684\u63a8\u7406\u8d26\u5355，\u5c31\u4f1a\u77e5\u9053\u5dee\u522b\u6709\u591a\u5927\u3002\u8fd9\u4e5f\u610f\u5473\u7740，\u6d4b\u8bd5\u4e00\u4e2a\u65b0\u6a21\u578b\u65f6\u4e0d\u9700\u8981\u91cd\u65b0\u8d70\u4e00\u5957\u72ec\u7acb\u91c7\u8d2d\u6d41\u7a0b\u3002<\/p>\n

\u8fd8\u6709\u4e00\u6761\u53ef\u4ee5\u79bb\u5f00\u6309 Token \u8ba1\u8d39\u7684\u8def\u5f84\u3002 \u672c\u6587\u6240\u6709\u6d4b\u8bd5\u90fd\u4f7f\u7528\u65e0\u670d\u52a1\u5668\u63a8\u7406\u3002\u5982\u679c\u4f60\u7684\u5de5\u4f5c\u8d1f\u8f7d\u7ee7\u7eed\u589e\u957f，\u4f8b\u5982\u9700\u8981\u6301\u7eed\u9ad8\u541e\u5410\u3001\u6d41\u91cf\u66f4\u52a0\u7a33\u5b9a\u53ef\u9884\u6d4b，\u6216\u8005\u5bf9\u8d44\u6e90\u9694\u79bb\u6709\u8981\u6c42，\u90a3\u4e48\u4e13\u5c5e\u63a8\u7406\u53ef\u4ee5\u7ee7\u7eed\u63d0\u4f9b\u76f8\u540c\u7684 OpenAI-compatible API，\u540c\u65f6\u4f7f\u7528\u9884\u7559 Capacity，\u56e0\u6b64\u8fc1\u79fb\u66f4\u50cf\u662f\u4e00\u6b21\u914d\u7f6e\u8c03\u6574，\u800c\u4e0d\u662f\u91cd\u5199\u5e94\u7528\u3002\u518d\u5f80\u4e0b\u4e00\u5c42，GPU Droplets \u5141\u8bb8\u4f60\u5728\u81ea\u5df1\u7684 GPU \u5b9e\u4f8b\u4e0a\u8fd0\u884c\u81ea\u5df1\u7684\u63a8\u7406\u670d\u52a1\u2014\u2014\u8fd9\u786e\u5b9e\u5c5e\u4e8e\u4e00\u6b21\u771f\u6b63\u7684\u67b6\u6784\u8c03\u6574，\u4f46\u4f9d\u7136\u53ef\u4ee5\u5728\u540c\u4e00\u4e2a\u8d26\u53f7\u4e0b\u5b8c\u6210，\u4e0d\u9700\u8981\u66f4\u6362 Cloud Provider\u3002<\/p>\n

\u8fd9\u4e9b\u56e0\u7d20\u90fd\u4e0d\u4f1a\u51fa\u73b0\u5728\u5355\u7eaf\u7684 Token \u5355\u4ef7\u6bd4\u8f83\u4e2d，\u800c\u8fd9\u6070\u6070\u662f\u91cd\u70b9\u3002\u5355 Token \u4ef7\u683c\u662f\u6700\u5bb9\u6613\u6bd4\u8f83\u7684\u6570\u5b57，\u4f46\u6b63\u5982\u672c\u6587\u524d\u9762\u7684\u6d4b\u8bd5\u6240\u5c55\u793a\u7684，\u5b83\u53cd\u800c\u6700\u4e0d\u53ef\u80fd\u5355\u72ec\u51b3\u5b9a\u4f60\u7684\u6700\u7ec8\u8d26\u5355\u3002<\/p>\n

\u6d4b\u8bd5\u65b9\u6cd5：\u8fd9\u4e9b\u6570\u5b57\u662f\u5982\u4f55\u5f97\u5230\u7684<\/h3>\n

\u6240\u6709\u6d4b\u8bd5\u90fd\u8fd0\u884c\u5728 DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u7684 https:\/\/inference.do-ai.run\/v1 API \u7aef\u70b9\u4e0a，\u8be5\u7aef\u70b9\u517c\u5bb9 OpenAI API\u3002<\/p>\n

\u6ca1\u6709\u53d1\u9001\u4efb\u4f55\u91c7\u6837\u53c2\u6570\u3002 \u6211\u4eec\u6ca1\u6709\u8bbe\u7f6e Temperature\u3001top_p \u6216 Penalty\u3002Kimi K3 \u4f7f\u7528\u56fa\u5b9a\u91c7\u6837\u53c2\u6570，\u56e0\u6b64\u5982\u679c\u53ea\u7ed9\u53e6\u5916\u4e24\u4e2a\u6a21\u578b\u8bbe\u7f6e\u8fd9\u4e9b\u53c2\u6570，\u5c31\u4f1a\u5f15\u5165\u65b0\u7684\u4e0d\u5bf9\u79f0，\u540e\u9762\u8fd8\u5f97\u989d\u5916\u89e3\u91ca\u3002\u4e09\u4e2a\u6a21\u578b\u5168\u90e8\u4f7f\u7528\u5e73\u53f0\u9ed8\u8ba4\u503c\u3002<\/p>\n

\u4f7f\u7528\u975e\u6d41\u5f0f\u8f93\u51fa，\u5e76\u8bbe\u7f6e max_tokens=32000\u3002 \u65e9\u671f\u4e00\u6b21\u6d4b\u8bd5\u4f7f\u7528 8000 Token \u4e0a\u9650，\u7ed3\u679c Flash \u7684 Tier 3 Response \u6709 20% \u88ab\u622a\u65ad\u2014\u2014\u4e5f\u5c31\u662f\u8bf4，Token Budget \u7528\u5b8c\u65f6\u6a21\u578b\u8fd8\u5728\u8fdb\u884c\u63a8\u7406，\u751a\u81f3\u8fd8\u6ca1\u8f93\u51fa\u6700\u7ec8\u7b54\u6848\u3002\u672c\u6587\u6700\u7ec8\u4f7f\u7528\u7684\u6570\u636e\u4e2d\u6ca1\u6709\u4efb\u4f55\u622a\u65ad\u8bb0\u5f55，\u6240\u6709 finish_reason \u90fd\u662f stop\u3002<\/p>\n

\u968f\u673a\u5316\u6267\u884c\u987a\u5e8f\u3002 \u5b8c\u6574\u7684\u201c\u6a21\u578b \u00d7 Prompt \u00d7 \u91cd\u590d\u6b21\u6570\u201d\u6d4b\u8bd5\u77e9\u9635\u4f7f\u7528\u56fa\u5b9a Seed \u6253\u4e71\u6267\u884c\u987a\u5e8f，\u8fd9\u6837\u5e73\u53f0\u968f\u7740\u65f6\u95f4\u4ea7\u751f\u7684\u6027\u80fd\u53d8\u5316\u4f1a\u5e73\u5747\u5206\u6563\u5230\u5404\u6a21\u578b，\u800c\u4e0d\u662f\u96c6\u4e2d\u5f71\u54cd\u6700\u540e\u6267\u884c\u7684\u90a3\u4e2a\u6a21\u578b\u3002<\/p>\n

Token \u7edf\u8ba1\u3002 DigitalOcean \u7684 usage Object \u4e0d\u4f1a\u5355\u72ec\u7edf\u8ba1\u63a8\u7406 Token，completion_tokens_details \u8fd4\u56de\u7684\u662f null\u3002\u4f46\u6a21\u578b\u4f1a\u628a\u63a8\u7406\u5185\u5bb9\u653e\u5728 Message \u7684 reasoning_content \u5b57\u6bb5\u4e2d\u3002\u6211\u4eec\u4f7f\u7528\u6bcf\u4e2a\u6a21\u578b\u81ea\u5df1\u7684 Tokenizer \u5728\u672c\u5730\u7edf\u8ba1\u8fd9\u90e8\u5206 Token，\u518d\u4e0e\u5b9e\u9645\u8ba1\u8d39\u7684 completion_tokens \u5bf9\u9f50\u3002\u5269\u4f59\u5dee\u503c\u662f\u6bcf\u4e2a\u6a21\u578b\u56fa\u5b9a\u7684 Chat Template Offset：Flash \u4e3a +2，Qwen \u4e3a +1，Kimi \u4e3a +13\u3002\u56fa\u5b9a Offset \u901a\u5e38\u8bf4\u660e Tokenizer \u4f7f\u7528\u6b63\u786e；\u5982\u679c\u8fd9\u4e2a\u5dee\u503c\u968f\u673a\u6ce2\u52a8，\u5c31\u610f\u5473\u7740\u6211\u4eec\u7684\u7edf\u8ba1\u65b9\u5f0f\u6709\u95ee\u9898\u3002\u672c\u6587\u6240\u6709\u6bd4\u4f8b\u90fd\u57fa\u4e8e\u5b9e\u9645\u8ba1\u8d39 Token，\u5e76\u4fdd\u7559\u8fd9\u4e9b Offset\u3002<\/p>\n

\u63a8\u7406\u63a7\u5236\u53c2\u6570\u901a\u8fc7\u4e09\u79cd\u4e0d\u540c\u65b9\u5f0f\u53d1\u9001，\u56e0\u4e3a\u5b83\u4eec\u4e0d\u80fd\u4e92\u6362\u3002enable_thinking \u548c clear_thinking \u653e\u5728 chat_template_kwargs \u4e2d，\u800c reasoning_effort \u653e\u5728 Request Body \u9876\u5c42\u3002\u6bcf\u4e00\u6b21\u8bf7\u6c42\u90fd\u4f1a\u8bb0\u5f55\u5b83\u5b9e\u9645\u53d1\u9001\u7684 extra_body\u3002\u6211\u4eec\u6ca1\u6709\u56e0\u4e3a API \u8fd4\u56de 200 \u5c31\u5047\u8bbe\u53c2\u6570\u771f\u7684\u751f\u6548，\u800c\u662f\u901a\u8fc7\u5b9e\u9645\u884c\u4e3a\u9a8c\u8bc1：\u4e0d\u540c\u6d4b\u8bd5\u7ec4\u4f1a\u9488\u5bf9\u540c\u6837\u7684 Prompt \u548c\u91cd\u590d\u6b21\u6570\u505a\u914d\u5bf9\u6bd4\u8f83\u3002\u5982\u679c\u53c2\u6570\u88ab\u5ffd\u7565，\u90a3\u4e48 Token \u5206\u5e03\u5e94\u8be5\u51e0\u4e4e\u6ca1\u6709\u5dee\u5f02\u3002<\/p>\n

\u4f7f\u7528\u4e2d\u4f4d\u6570，\u800c\u4e0d\u662f\u5e73\u5747\u503c，\u539f\u56e0\u524d\u9762\u5df2\u7ecf\u89e3\u91ca\u8fc7\u30025 \u6b21\u91cd\u590d\u6d4b\u8bd5\u8db3\u4ee5\u652f\u6301\u4e00\u4e2a\u4e2d\u4f4d\u6570\u548c\u8303\u56f4，\u4f46\u4e0d\u8db3\u4ee5\u7ed9\u51fa\u975e\u5e38\u7a84\u7684\u7f6e\u4fe1\u533a\u95f4\u3002<\/p>\n

\u8fd9\u9879\u7814\u7a76\u7684\u9650\u5236<\/h3>\n

\u672c\u6587\u6d4b\u7684\u662f\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u548c\u5355\u4e2a\u6b63\u786e\u7b54\u6848\u6210\u672c，\u800c\u4e0d\u662f\u5b8c\u6210\u6574\u4e2a\u4efb\u52a1\u7684\u6210\u672c\u3002 \u6ca1\u6709 Sandbox\u3001\u591a\u8f6e Agent Loop，\u4e5f\u6ca1\u6709\u4ee3\u7801\u6267\u884c\u3002Agent \u5de5\u4f5c\u8d1f\u8f7d\u663e\u7136\u662f\u4e0b\u4e00\u6b65\u6700\u503c\u5f97\u6d4b\u7684\u65b9\u5411，\u4f46\u6211\u4eec\u76ee\u524d\u8fd8\u6ca1\u6709\u505a\u3002<\/p>\n

\u6a21\u578b\u80fd\u529b\u901a\u8fc7\u7b2c\u4e09\u65b9\u6570\u636e\u4fdd\u6301\u53ef\u6bd4，\u800c\u4e0d\u662f\u7531\u6211\u4eec\u91cd\u65b0\u6d4b\u8bd5\u3002 Artificial Analysis \u5728 Intelligence Index \u4e0a\u7ed9 Kimi K3 \u548c GLM-5.3-Flash \u90fd\u6253\u4e86 57 \u5206\u3002\u8fd9\u662f\u4ed6\u4eec\u7684\u6d4b\u91cf\u7ed3\u679c，\u4e0d\u662f\u6211\u4eec\u7684，\u6211\u4eec\u4e5f\u6ca1\u6709\u5c1d\u8bd5\u590d\u73b0\u3002<\/p>\n

\u53ef\u9a8c\u8bc1\u4efb\u52a1\u96c6\u592a\u7b80\u5355\u3002 11 \u79cd\u914d\u7f6e\u91cc\u6709 6 \u79cd\u62ff\u5230\u4e86 100%\u3002\u8fd9\u5957\u6d4b\u8bd5\u8db3\u4ee5\u53d1\u73b0\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u4f1a\u8ba9\u51c6\u786e\u7387\u4e0b\u964d\u2014\u2014low \u4e0b\u662f 92%\u2014\u2014\u4f46\u4e0d\u8db3\u4ee5\u544a\u8bc9\u4f60\u5728\u771f\u6b63\u56f0\u96be\u7684\u5de5\u4f5c\u8d1f\u8f7d\u4e2d\u4f1a\u4e0b\u964d\u591a\u5c11\u3002\u56e0\u6b64，\u8fd9\u91cc\u7684\u51c6\u786e\u7387\u6bd4\u8f83\u53ea\u80fd\u8bf4\u660e：\u5728\u8fd9\u4e00\u96be\u5ea6\u3001\u8fd9\u4e00\u7c7b\u4efb\u52a1\u4e0a，\u6a21\u578b\u8868\u73b0\u63a5\u8fd1，\u4e0d\u80fd\u63a8\u5e7f\u5230\u66f4\u5e7f\u6cdb\u573a\u666f\u3002<\/p>\n

\u6211\u4eec\u4e00\u5f00\u59cb\u6d4b\u8bd5\u9519\u4e86\u53c2\u6570\u3002 \u6700\u521d\u7684\u5b9e\u9a8c\u4f7f\u7528\u4e86 enable_thinking，\u4f46\u8fd9\u5e76\u4e0d\u662f GLM-5.3-Flash Model Card \u63a8\u8350\u7684\u63a7\u5236\u65b9\u5f0f\u3002\u76f4\u5230\u6211\u4eec\u91cd\u65b0\u4ed4\u7ec6\u9605\u8bfb Model Card \u540e，\u624d\u53d1\u73b0\u771f\u6b63\u5e94\u8be5\u4f7f\u7528\u7684\u662f reasoning_effort，\u800c\u8fd9\u4e2a\u53c2\u6570\u663e\u8457\u6539\u53d8\u4e86\u7ed3\u8bba\u3002\u672c\u6587\u4ecd\u7136\u4fdd\u7559 enable_thinking \u7684\u6d4b\u8bd5\u7ed3\u679c，\u4f46\u5e94\u8be5\u628a\u5b83\u7406\u89e3\u4e3a\u201c\u8fd9\u4e2a\u53c2\u6570\u5728\u6a21\u578b\u4e0a\u7684\u884c\u4e3a\u201d，\u800c\u4e0d\u662f\u201c\u6a21\u578b\u80fd\u4e0d\u80fd\u51cf\u5c11\u63a8\u7406\u201d\u7684\u6700\u7ec8\u7ed3\u8bba\u3002<\/p>\n

\u6211\u4eec\u7684\u7b54\u6848 Key \u6709\u4e00\u9053\u9898\u5199\u9519\u4e86\u3002 \u5728\u5ef6\u8fdf\u9884\u7b97\u4efb\u52a1\u4e2d，45 \u6761 Response \u5168\u90e8\u56de\u7b54 43，\u800c\u6211\u4eec\u7684\u7b54\u6848 Key \u5199\u7684\u662f 42\u3002\u540e\u6765\u786e\u8ba4\u6a21\u578b\u662f\u5bf9\u7684：\u7b2c\u4e00\u4e2a Token \u5728 TTFT \u65f6\u5df2\u7ecf\u5230\u8fbe，\u53ea\u6709\u540e\u7eed Token \u624d\u4f1a\u7ee7\u7eed\u589e\u52a0\u603b\u5ef6\u8fdf\u3002\u6211\u4eec\u4fee\u6b63\u4e86\u7b54\u6848 Key，\u5e76\u4e14\u9009\u62e9\u516c\u5f00\u8fd9\u4e2a\u95ee\u9898，\u56e0\u4e3a\u4e00\u4e2a\u6c38\u8fdc\u4e0d\u4f1a\u51fa\u9519\u7684\u56fa\u5b9a\u7b54\u6848 Key，\u5f88\u591a\u65f6\u5019\u53ea\u662f\u56e0\u4e3a\u6839\u672c\u6ca1\u4eba\u771f\u6b63\u68c0\u67e5\u8fc7\u3002<\/p>\n

Kimi K3 \u6709 12 \u6761 Response \u5b58\u5728\u65e0\u6cd5\u89e3\u91ca\u7684\u8ba1\u8d39 Token\u3002 \u8fd9\u4e9b\u8bb0\u5f55\u91cc\u7684 completion_tokens \u6bd4\u5b9e\u9645\u8fd4\u56de\u7684\u63a8\u7406\u5185\u5bb9 + \u6700\u7ec8\u7b54\u6848\u591a\u51fa 85～391 Token\u3002\u5728\u5927\u7ea6 1000 \u6761 Kimi Record \u4e2d\u5171\u6709 12 \u6761\u51fa\u73b0\u8fd9\u79cd\u60c5\u51b5，\u6211\u4eec\u6ca1\u80fd\u89e3\u91ca\u539f\u56e0\u3002Flash \u548c Qwen \u6ca1\u6709\u51fa\u73b0\u7c7b\u4f3c\u5c3e\u90e8\u73b0\u8c61\u3002<\/p>\n

2700 \u6b21\u8c03\u7528\u91cc\u6709 1 \u6b21\u5931\u8d25\u3002 \u8fd9\u6b21\u8bf7\u6c42\u5728\u91cd\u8bd5 5 \u6b21\u4e4b\u540e\u4ecd\u7136\u8fd4\u56de HTTP 429\u3002<\/p>\n

297 \u6761\u8bb0\u5f55\u89e6\u53d1\u4e86 Prompt Cache\u3002 \u6bcf\u6b21\u90fd\u6070\u597d\u7f13\u5b58 128 Token，\u56e0\u4e3a\u76f8\u540c Prompt \u5728\u540e\u7eed\u91cd\u590d\u6d4b\u8bd5\u4e2d\u547d\u4e2d\u4e86 Cache\u3002\u8fd9\u53ea\u4f1a\u5f71\u54cd\u8f93\u5165\u6210\u672c，\u800c\u8f93\u5165\u6210\u672c\u5728\u8fd9\u7ec4\u8d26\u5355\u91cc\u5360\u6bd4\u5f88\u5c0f\u3002\u6211\u4eec\u6ca1\u6709\u4eba\u4e3a\u7981\u7528\u5b83，\u800c\u662f\u4fdd\u7559\u4e86\u771f\u5b9e\u884c\u4e3a\u3002<\/p>\n

\u4e24\u5957 Prompt Set \u7684 Token \u6570\u4e0d\u80fd\u76f4\u63a5\u4e92\u76f8\u6bd4\u3002 \u53ef\u9a8c\u8bc1\u4efb\u52a1\u989d\u5916\u5e26\u6709\u683c\u5f0f\u8bf4\u660e，\u800c\u5f00\u653e\u5f0f Prompt \u6ca1\u6709\u3002\u5e94\u8be5\u6bd4\u8f83\u540c\u4e00\u4e2a\u4efb\u52a1\u96c6\u5185\u4e0d\u540c\u6a21\u578b\u4e4b\u95f4\u7684\u6bd4\u4f8b，\u800c\u4e0d\u662f\u8de8\u4e24\u4e2a\u4efb\u52a1\u96c6\u6bd4\u8f83\u7edd\u5bf9 Token \u6570\u3002<\/p>\n

\u6240\u6709\u6570\u636e\u90fd\u53ea\u662f\u67d0\u4e00\u65f6\u95f4\u70b9\u7684\u7ed3\u679c\u3002 \u4ef7\u683c\u3001Serving Default \u548c\u6a21\u578b\u7248\u672c\u90fd\u4f1a\u53d8\u5316\u3002\u4ee3\u7801\u4ed3\u5e93\u91cc\u8bb0\u5f55\u4e86\u672c\u6b21\u6d4b\u8bd5\u6240\u4f7f\u7528\u7684\u4ef7\u683c\u5feb\u7167\u65e5\u671f\u3002<\/p>\n

\u81ea\u5df1\u590d\u73b0\u8fd9\u4e9b\u6d4b\u8bd5<\/h3>\n

\u4ee3\u7801\u4ed3\u5e93(github.com\/Jameshskelton\/GLM-Flash-Analysis)\u4e2d\u5305\u542b\u56fa\u5b9a\u7248\u672c\u7684 Prompt Set \u53ca\u5176 SHA256\u3001\u6d4b\u8bd5\u811a\u672c\u3001\u6240\u6709\u539f\u59cb Response Log\u3001Grader，\u4ee5\u53ca\u91cd\u65b0\u751f\u6210\u672c\u6587\u5168\u90e8\u6570\u5b57\u548c\u56fe\u8868\u7684\u5206\u6790\u811a\u672c\u3002\u4f7f\u7528\u7684\u6a21\u578b ID \u5206\u522b\u662f glm-5.3-flash\u3001qwen3.8-max \u548c kimi-k3\u3002<\/p>\n

\u6574\u4e2a\u7814\u7a76\u5927\u7ea6\u53ea\u9700\u8981 14 \u7f8e\u5143\u548c\u534a\u5929\u65f6\u95f4\u3002<\/p>\n

\u5982\u679c\u4f60\u91cd\u65b0\u8fd0\u884c\u540e\u5f97\u5230\u4e0d\u540c\u7684\u6570\u636e，\u6b22\u8fce\u63d0\u4ea4 Issue\u3002\u6211\u4eec\u4e5f\u5f88\u60f3\u77e5\u9053\u7ed3\u679c\u3002<\/p>\n

FAQ<\/h3>\n

GLM-5.3-Flash \u662f DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u4e0a\u6700\u4fbf\u5b9c\u7684\u6a21\u578b\u5417？<\/p>\n

\u5728\u6211\u4eec\u7684\u6d4b\u8bd5\u8303\u56f4\u5185，\u662f\u7684，\u800c\u4e14\u5728\u6240\u6709\u6d4b\u8bd5\u914d\u7f6e\u4e2d\u90fd\u662f\u6700\u4fbf\u5b9c\u7684\u3002\u9ed8\u8ba4\u8bbe\u7f6e\u4e0b，\u5b83\u7684\u5355\u6b21\u8bf7\u6c42\u6210\u672c\u662f 0.00156 \u7f8e\u5143，Qwen3.8-Max \u4e3a 0.00361 \u7f8e\u5143，Kimi K3 \u4e3a 0.02661 \u7f8e\u5143\u3002\u8bbe\u7f6e reasoning_effort: low \u540e，\u6210\u672c\u8fdb\u4e00\u6b65\u964d\u5230 0.00025 \u7f8e\u5143\u2014\u2014\u6bd4 Qwen \u4fbf\u5b9c 14.5 \u500d，\u6bd4 Kimi \u4fbf\u5b9c 106 \u500d\u3002<\/p>\n

\u5173\u95ed\u601d\u8003\u6a21\u5f0f\u80fd\u964d\u4f4e\u6210\u672c\u5417？<\/p>\n

\u53d6\u51b3\u4e8e\u6a21\u578b，\u4e5f\u53d6\u51b3\u4e8e\u4f60\u4f7f\u7528\u7684\u662f\u54ea\u4e2a\u53c2\u6570\u3002\u5728 Kimi K3 \u4e0a，enable_thinking: false \u53ef\u4ee5\u964d\u4f4e 75.5% \u7684\u6210\u672c\u3002Qwen3.8-Max \u9ed8\u8ba4\u5df2\u7ecf\u5173\u95ed\u601d\u8003\u6a21\u5f0f\u3002\u800c\u5728 GLM-5.3-Flash \u4e0a，enable_thinking: false \u53cd\u800c\u8ba9\u6210\u672c\u589e\u52a0\u4e86 11.1%，\u56e0\u4e3a\u6a21\u578b\u7ee7\u7eed\u5728\u7528\u6237\u53ef\u89c1\u7b54\u6848\u91cc\u8fdb\u884c\u63a8\u7406；\u771f\u6b63\u6709\u6548\u7684\u53c2\u6570\u662f reasoning_effort，\u6700\u591a\u53ef\u4ee5\u964d\u4f4e 84% \u6210\u672c\u3002<\/p>\n

GLM-5.3-Flash \u9ed8\u8ba4\u7684 reasoning_effort \u662f\u4ec0\u4e48？<\/p>\n

max\u3002\u5982\u679c\u8bf7\u6c42\u91cc\u6ca1\u6709\u53d1\u9001 reasoning_effort，\u6a21\u578b\u5c31\u4f1a\u8fd0\u884c\u5728 max\u3002\u6211\u4eec\u9488\u5bf9\u5b8c\u5168\u76f8\u540c\u7684 Prompt，\u628a\u672a\u8bbe\u7f6e\u548c\u663e\u5f0f max \u505a\u4e86\u5bf9\u6bd4，\u4e2d\u4f4d\u6bd4\u503c\u662f 0.984\u3002\u8fd9\u610f\u5473\u7740\u6bcf\u4e2a\u672a\u663e\u5f0f\u914d\u7f6e\u7684\u8bf7\u6c42，\u5b9e\u9645\u4e0a\u90fd\u8fd0\u884c\u5728\u6700\u6602\u8d35\u7684\u8bbe\u7f6e\u3002<\/p>\n

\u964d\u4f4e\u63a8\u7406\u5f3a\u5ea6\u4f1a\u5f71\u54cd\u51c6\u786e\u7387\u5417？<\/p>\n

\u4f1a，\u4e3b\u8981\u51fa\u73b0\u5728\u56f0\u96be\u95ee\u9898\u4e0a\u3002\u5728 20 \u4e2a\u53ef\u9a8c\u8bc1\u4efb\u52a1\u4e2d，GLM-5.3-Flash \u9ed8\u8ba4\u914d\u7f6e\u5f97\u5206 100%，high \u4e3a 98%，low \u4e3a 92%\u3002\u6240\u6709\u5931\u8d25\u90fd\u53d1\u751f\u5728\u591a\u6b65\u63a8\u7406\u4efb\u52a1\u4e0a，\u7b80\u5355\u4efb\u52a1\u5728\u6240\u6709\u7b49\u7ea7\u4e0b\u90fd\u6ca1\u6709\u53d7\u5230\u5f71\u54cd\u3002<\/p>\n

\u4e3a\u4ec0\u4e48\u6211\u7684 LLM \u8d26\u5355\u6bd4\u4ef7\u683c\u9875\u9762\u7b97\u51fa\u6765\u7684\u9ad8？<\/p>\n

\u6700\u53ef\u80fd\u7684\u539f\u56e0\u662f\u8f93\u51fa Token \u6570\u91cf\u3002\u63a8\u7406\u6a21\u578b\u4f1a\u751f\u6210\u5927\u91cf\u8ba1\u8d39\u4f46\u4e0d\u4f1a\u5c55\u793a\u7ed9\u7528\u6237\u7684 Token\u3002\u5728\u6211\u4eec\u7684\u6d4b\u8bd5\u4e2d，\u8fd9\u90e8\u5206\u5360\u8f93\u51fa\u8d39\u7528\u7684 65%～84%\u3002\u5355 Token \u4ef7\u683c\u53ea\u544a\u8bc9\u4f60\u5355\u4ef7，\u5374\u4e0d\u4f1a\u544a\u8bc9\u4f60\u6700\u7ec8\u4f1a\u751f\u6210\u591a\u5c11 Token\u3002<\/p>\n

LLM \u8d26\u5355\u91cc\u63a8\u7406 Token \u5360\u591a\u5c11？<\/p>\n

GLM-5.3-Flash \u9ed8\u8ba4\u8bbe\u7f6e\u4e0b，\u7b80\u5355\u95ee\u9898\u4e2d\u4f4d\u6570\u4e3a 72%，\u56f0\u96be\u95ee\u9898\u4f1a\u4e0a\u5347\u5230 84%\u3002\u4f7f\u7528 reasoning_effort: low \u540e，\u7b80\u5355\u95ee\u9898\u964d\u5230 3%，\u56f0\u96be\u95ee\u9898\u7ea6\u4e3a 40%\u3002Kimi K3 \u5219\u4ece 65% \u4e0a\u5347\u5230 82%\u3002Qwen3.8-Max \u9ed8\u8ba4\u5173\u95ed\u601d\u8003\u6a21\u5f0f，\u56e0\u6b64\u9ed8\u8ba4\u4e0d\u4f1a\u627f\u62c5\u8fd9\u90e8\u5206\u989d\u5916\u63a8\u7406\u6210\u672c\u3002<\/p>\n

\u4fbf\u5b9c\u6a21\u578b\u7684\u7b54\u6848\u4f1a\u66f4\u5dee\u5417？<\/p>\n

\u5728\u9ed8\u8ba4\u914d\u7f6e\u4e0b\u6ca1\u6709\u3002\u6211\u4eec\u4e00\u5171\u8bc4\u5206\u4e86 899 \u6761 Response，\u8986\u76d6 3 \u4e2a\u6a21\u578b\u548c\u5404\u81ea 3 \u79cd enable_thinking \u6a21\u5f0f\u3002\u5728\u6709\u786e\u5b9a\u7b54\u6848\u7684\u95ee\u9898\u4e0a，\u4e09\u4e2a\u6a21\u578b\u90fd\u8fbe\u5230\u6216\u63a5\u8fd1 100%；\u6700\u4f4e\u7684\u662f\u5173\u95ed\u601d\u8003\u6a21\u5f0f\u7684 Kimi K3，\u4e3a 98%\u3002\u4f46\u5355\u4e2a\u6b63\u786e\u7b54\u6848\u6210\u672c\u76f8\u5dee\u8fbe\u5230 33 \u500d\u3002\u964d\u4f4e GLM-5.3-Flash \u7684\u63a8\u7406\u5f3a\u5ea6\u786e\u5b9e\u4f1a\u5f71\u54cd\u51c6\u786e\u7387：high \u4e3a 98%，low \u4e3a 92%\u3002\u56e0\u6b64\u5b8c\u6574\u8bc4\u5206\u6570\u636e\u4e00\u5171\u5305\u542b 11 \u79cd\u914d\u7f6e\u30011099 \u6761 Response\u3002<\/p>\n

\u5728\u54ea\u91cc\u53ef\u4ee5\u8fd0\u884c GLM-5.3-Flash？<\/p>\n

\u6a21\u578b\u6743\u91cd\u91c7\u7528 MIT \u8bb8\u53ef\u8bc1，\u5e76\u5df2\u7ecf\u53d1\u5e03\u5728 Hugging Face，\u56e0\u6b64\u65e2\u53ef\u4ee5\u81ea\u884c\u90e8\u7f72，\u4e5f\u53ef\u4ee5\u901a\u8fc7\u4efb\u4f55\u63d0\u4f9b\u8be5\u6a21\u578b\u7684\u5e73\u53f0\u8fd0\u884c\u3002\u6211\u4eec\u662f\u5728 DigitalOcean \u65e0\u670d\u52a1\u5668\u63a8\u7406\u4e0a\u8fdb\u884c\u6d4b\u8bd5\u7684，\u76ee\u524d\u5b83\u4e0e Qwen3.8-Max\u3001Kimi K3 \u90fd\u53ef\u4ee5\u901a\u8fc7 https:\/\/inference.do-ai.run\/v1 \u8fd9\u4e2a\u517c\u5bb9 OpenAI \u7684 API \u7aef\u70b9\u8c03\u7528，GLM-5.3-Flash \u7684\u6a21\u578b ID \u662f glm-5.3-flash\u3002<\/p>\n

GLM-5.3-Flash \u5728 DigitalOcean \u4e0a\u6bd4\u5176\u4ed6\u5e73\u53f0\u66f4\u4fbf\u5b9c\u5417？<\/p>\n

\u5355 Token \u4ef7\u683c\u5e76\u4e0d\u662f\u5dee\u5f02\u70b9：\u8fd9\u4e2a\u6a21\u578b\u662f\u5f00\u653e\u6743\u91cd\u6a21\u578b，\u800c\u4e14 DigitalOcean \u7684\u4ef7\u683c\u4e0e Z.ai \u516c\u5e03\u7684\u5b98\u65b9\u4ef7\u683c\u4e00\u81f4\u3002\u771f\u6b63\u4f1a\u5f71\u54cd\u603b\u6210\u672c\u7684，\u662f\u6a21\u578b\u4e0e\u5176\u4ed6\u6280\u672f\u6808\u4e4b\u95f4\u7684\u8ddd\u79bb\u2014\u2014\u4f8b\u5982\u6bcf\u6b21 Retrieval \u4ea7\u751f\u7684\u51fa\u7ad9\u6d41\u91cf\u548c\u7f51\u7edc\u5f80\u8fd4\u5ef6\u8fdf\u5e76\u4e0d\u4f1a\u4f53\u73b0\u5728\u5355 Token \u4ef7\u683c\u6bd4\u8f83\u4e2d\u2014\u2014\u4ee5\u53ca\u5f53\u5de5\u4f5c\u8d1f\u8f7d\u589e\u957f\u540e，\u80fd\u4e0d\u80fd\u5728\u4e0d\u91cd\u65b0\u8fc1\u79fb\u6574\u5957\u5e73\u53f0\u7684\u60c5\u51b5\u4e0b\u5207\u6362\u5230\u4e13\u5c5e Capacity\u3002<\/p>\n

\u5982\u4f55\u964d\u4f4e\u8f93\u51fa Token \u6210\u672c？<\/p>\n

\u6709\u4e09\u4e2a\u4e3b\u8981\u624b\u6bb5，\u6309\u5f71\u54cd\u5927\u5c0f\u6392\u5e8f\u3002\u7b2c\u4e00，\u5728 GLM-5.3-Flash \u4e0a\u663e\u5f0f\u8bbe\u7f6e reasoning_effort：low \u6bd4\u9ed8\u8ba4\u4fbf\u5b9c 6.3 \u500d，high \u4fbf\u5b9c 4 \u500d\u3002\u7b2c\u4e8c，\u9650\u5236\u8f93\u51fa\u683c\u5f0f：\u5f3a\u5236\u4f7f\u7528\u56fa\u5b9a\u7b54\u6848\u7ed3\u6784，\u8ba9 Flash \u7684\u8f93\u51fa Token \u4e2d\u4f4d\u6570\u964d\u4f4e\u4e86 80.4%\u3002\u7b2c\u4e09，\u52a0\u4e0a\u7b80\u6d01\u56de\u7b54\u8981\u6c42，\u4f8b\u5982 \u201cAnswer in one sentence\u201d，\u53ef\u4ee5\u8ba9\u4e09\u4e2a\u6a21\u578b\u5728\u7b80\u5355\u95ee\u9898\u4e0a\u7684\u8f93\u51fa\u51cf\u5c11 70%～91%\u3002\u5176\u4e2d\u53ea\u6709\u7b2c\u4e00\u4e2a\u65b9\u6cd5\u771f\u6b63\u964d\u4f4e\u4e86\u8d26\u5355\u91cc\u63a8\u7406 Token \u6240\u5360\u7684\u6bd4\u4f8b；\u53e6\u5916\u4e24\u4e2a\u65b9\u6cd5\u53ea\u662f\u8ba9\u603b\u8d26\u5355\u53d8\u5c0f，\u800c\u63a8\u7406 Token \u5360\u6bd4\u57fa\u672c\u4e0d\u53d8\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u6211\u4eec\u95ee\u4e86 GLM-5.3-Flash \u4e00\u4e2a\u975e\u5e38\u7b80\u5355\u7684\u95ee\u9898：SSH \u9ed8\u8ba4\u4f7f\u7528\u54ea\u4e2a\u7aef\u53e3\u3002\u5b83\u7528\u4e86 625 \u4e2a\u8f93\u51fa Token，\u6700\u540e\u624d\u7ed9\u51fa\u7b54\u6848：\u201c22\u201d\u3002
\n\u8fd9\u4e0d\u662f\u6211\u4eec\u7279\u610f\u6311\u51fa\u6765\u7684\u6781\u7aef\u6848\u4f8b，\u800c\u662f 5 \u6b21\u6d4b\u8bd5\u7684\u4e2d\u4f4d\u6570\u3002Qwen3.8-Max \u56de\u7b54\u540c\u4e00\u4e2a\u95ee\u9898\u53ea\u7528\u4e86 156 \u4e2a Token\u3002Flash \u8fd9\u4e9b\u8c03\u7528\u4e2d，\u6211\u4eec\u652f\u4ed8\u7684\u5927\u90e8\u5206\u8f93\u51fa\u6210\u672c，\u5176\u5b9e\u90fd\u82b1\u5728\u4e86\u7528\u6237\u6839\u672c\u770b\u4e0d\u5230\u7684\u63a8\u7406\u8fc7\u7a0b\u4e0a，\u800c\u8fd9\u4e2a\u95ee\u9898\u6700\u7ec8<\/p>\n","protected":false},"author":2,"featured_media":103501,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[74,53,75,50],"topic":[],"class_list":["post-103505","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-agent","tag-ai","tag-llm","tag-50"],"yoast_head":"\nGLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \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\/103505.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6211\u4eec\u95ee\u4e86 GLM-5.3-Flash \u4e00\u4e2a\u975e\u5e38\u7b80\u5355\u7684\u95ee\u9898：SSH \u9ed8\u8ba4\u4f7f\u7528\u54ea\u4e2a\u7aef\u53e3\u3002\u5b83\u7528\u4e86 625 \u4e2a\u8f93\u51fa Token，\u6700\u540e\u624d\u7ed9\u51fa\u7b54\u6848：\u201c22\u201d\u3002 \u8fd9\u4e0d\u662f\u6211\u4eec\u7279\u610f\u6311\u51fa\u6765\u7684\u6781\u7aef\u6848\u4f8b，\u800c\u662f 5 \u6b21\u6d4b\u8bd5\u7684\u4e2d\u4f4d\u6570\u3002Qwen3.8-Max \u56de\u7b54\u540c\u4e00\u4e2a\u95ee\u9898\u53ea\u7528\u4e86 156 \u4e2a Token\u3002Flash \u8fd9\u4e9b\u8c03\u7528\u4e2d，\u6211\u4eec\u652f\u4ed8\u7684\u5927\u90e8\u5206\u8f93\u51fa\u6210\u672c，\u5176\u5b9e\u90fd\u82b1\u5728\u4e86\u7528\u6237\u6839\u672c\u770b\u4e0d\u5230\u7684\u63a8\u7406\u8fc7\u7a0b\u4e0a，\u800c\u8fd9\u4e2a\u95ee\u9898\u6700\u7ec8\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/103505.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-10T11:17:40+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/09\/20260910111734-6aa291ce2a6cb.png\" \/>\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=\"8 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/103505.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/103505.html\",\"name\":\"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-09-10T11:17:40+00:00\",\"dateModified\":\"2026-09-10T11:17:40+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/103505.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/103505.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/103505.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d\"}]},{\"@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":"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \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\/103505.html","og_locale":"zh_CN","og_type":"article","og_title":"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6211\u4eec\u95ee\u4e86 GLM-5.3-Flash \u4e00\u4e2a\u975e\u5e38\u7b80\u5355\u7684\u95ee\u9898：SSH \u9ed8\u8ba4\u4f7f\u7528\u54ea\u4e2a\u7aef\u53e3\u3002\u5b83\u7528\u4e86 625 \u4e2a\u8f93\u51fa Token，\u6700\u540e\u624d\u7ed9\u51fa\u7b54\u6848：\u201c22\u201d\u3002 \u8fd9\u4e0d\u662f\u6211\u4eec\u7279\u610f\u6311\u51fa\u6765\u7684\u6781\u7aef\u6848\u4f8b，\u800c\u662f 5 \u6b21\u6d4b\u8bd5\u7684\u4e2d\u4f4d\u6570\u3002Qwen3.8-Max \u56de\u7b54\u540c\u4e00\u4e2a\u95ee\u9898\u53ea\u7528\u4e86 156 \u4e2a Token\u3002Flash \u8fd9\u4e9b\u8c03\u7528\u4e2d，\u6211\u4eec\u652f\u4ed8\u7684\u5927\u90e8\u5206\u8f93\u51fa\u6210\u672c，\u5176\u5b9e\u90fd\u82b1\u5728\u4e86\u7528\u6237\u6839\u672c\u770b\u4e0d\u5230\u7684\u63a8\u7406\u8fc7\u7a0b\u4e0a，\u800c\u8fd9\u4e2a\u95ee\u9898\u6700\u7ec8","og_url":"https:\/\/www.wsisp.com\/helps\/103505.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-09-10T11:17:40+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/09\/20260910111734-6aa291ce2a6cb.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"8 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/103505.html","url":"https:\/\/www.wsisp.com\/helps\/103505.html","name":"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-09-10T11:17:40+00:00","dateModified":"2026-09-10T11:17:40+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/103505.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/103505.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/103505.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"GLM-5.3-Flash \u6210\u672c\u4f18\u5316\u5b9e\u6218\uff1a\u52a0\u4e00\u4e2a\u53c2\u6570\uff0c\u63a8\u7406\u6210\u672c\u76f4\u964d 6.3 \u500d"}]},{"@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\/103505","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=103505"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/103505\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/103501"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=103505"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=103505"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=103505"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=103505"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}