{"id":1228,"date":"2024-11-19T11:10:28","date_gmt":"2024-11-19T03:10:28","guid":{"rendered":"https:\/\/aws-oncloudai.com\/?p=1228"},"modified":"2024-11-20T10:03:53","modified_gmt":"2024-11-20T02:03:53","slug":"updates-to-claude-sonnet-3-5-and-claude-3-5-haiku","status":"publish","type":"post","link":"https:\/\/aws-oncloudai.com\/zh_hk\/updates-to-claude-sonnet-3-5-and-claude-3-5-haiku\/","title":{"rendered":"Claude Sonnet 3.5 \u548cClaude 3.5 Haiku \u7684\u66f4\u65b0"},"content":{"rendered":"<p>\u6211\u5011<strong>Oncloud AI<\/strong>\u900f\u904e\u672c\u6587\u8a73\u7d30\u4e86\u89e3Claude Sonnet 3.5 \u548cClaude 3.5 Haiku \u7684\u6700\u65b0\u529f\u80fd\u548c\u589e\u5f37\u529f\u80fd\uff0c\u5305\u62ec\u6539\u9032\u7684\u6548\u80fd\u3001\u65b0\u529f\u80fd\u548c\u4f7f\u7528\u8005\u53cb\u597d\u7684\u66f4\u65b0\u3002\u96a8\u6642\u4e86\u89e3\u9019\u4e9b\u7248\u672c\u5728\u9ad8\u968eAI \u8a69\u6b4c\u5de5\u5177\u9818\u57df\u7684\u7368\u7279\u4e4b\u8655\u3002<\/p>\n<h3 id=\"heading-performance-improvements\" class=\"permalink-heading\">\u6027\u80fd\u6539\u9032<\/h3>\n<h4><strong>\u7de8\u78bc\u80fd\u529b<\/strong><\/h4>\n<ul>\n<li>SWE-bench Verified \u5f97\u5206\u5f9e33.4% \u63d0\u9ad8\u523049.0%\uff0c\u8d85\u8d8a\u5176\u4ed6\u516c\u958b\u6a21\u578b<\/li>\n<li>\u589e\u5f37\u4ee3\u7406\u5de5\u5177\u4f7f\u7528\u4efb\u52d9(TAU-bench) \u7684\u6548\u80fd\uff1a\n<ul>\n<li>\u96f6\u552e\u9818\u57df\uff1a\u5f9e62.6% \u63d0\u9ad8\u523069.2%<\/li>\n<li>\u822a\u7a7a\u9818\u57df\uff1a\u753136.0%\u589e\u81f346.0%<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h4><strong>\u901f\u5ea6\u8207\u6548\u7387<\/strong><\/h4>\n<ul>\n<li>\u904b\u884c\u901f\u5ea6\u662fClaude 3 Opus \u7684\u5169\u500d<\/li>\n<li>\u5118\u7ba1\u6709\u6240\u6539\u9032\uff0c\u4f46\u6210\u672c\u7d50\u69cb\u4fdd\u6301\u4e0d\u8b8a<\/li>\n<\/ul>\n<h3 id=\"heading-new-features\" class=\"permalink-heading\">\u65b0\u529f\u80fd<\/h3>\n<h4><strong>\u96fb\u8166\u4f7f\u7528\uff08\u516c\u958b\u6e2c\u8a66\u7248\uff09<\/strong><\/h4>\n<ul>\n<li>\u5141\u8a31Claude \u50cf\u4eba\u985e\u4e00\u6a23\u8207\u96fb\u8166\u4ecb\u9762\u9032\u884c\u4ea4\u4e92<\/li>\n<li>\u53ef\u700f\u89bd\u87a2\u5e55\u3001\u884c\u52d5\u904a\u6a19\u548c\u8f38\u5165\u6587\u5b57<\/li>\n<li>OSWorld \u57fa\u6e96\u6e2c\u8a66\u5f97\u5206\u70ba14.9%\uff0c\u986f\u8457\u9ad8\u65bc\u7af6\u722d\u5c0d\u624b\u76847.7%<\/li>\n<\/ul>\n<h4><strong>\u6587\u7269\u7279\u5fb5<\/strong><\/h4>\n<ul>\n<li>\u5728\u5c0d\u8a71\u65c1\u908a\u5efa\u7acb\u5c08\u7528\u8996\u7a97\u4f86\u986f\u793a\u7522\u751f\u7684\u5167\u5bb9<\/li>\n<li>\u652f\u63f4\u4e09\u7a2e\u985e\u578b\u7684\u5de5\u4ef6\uff1a\n<ul>\n<li>\u57fa\u65bc\u6587\u5b57\u7684\u5beb\u4f5c\u4efb\u52d9<\/li>\n<li>\u9069\u7528\u65bc\u9700\u8981\u8996\u89ba\u6548\u679c\u7684\u9805\u76ee<\/li>\n<li>\u958b\u767c\u5de5\u4f5c\u7de8\u78bc<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h3 id=\"heading-model-variants\" class=\"permalink-heading\">\u578b\u865f\u8b8a\u9ad4<\/h3>\n<p><strong>Claude 3.5 Sonnet<\/strong><\/p>\n<ul>\n<li>\u73fe\u5df2\u63a8\u51fa\uff0c\u5404\u9805\u6307\u6a19\u5747\u589e\u5f37<\/li>\n<li>\u5177\u6709\u51fa\u8272\u7684\u7814\u7a76\u751f\u7a0b\u5ea6\u63a8\u7406\u80fd\u529b\u548c\u672c\u79d1\u751f\u7a0b\u5ea6\u7684\u77e5\u8b58<\/li>\n<li>\u6539\u9032\u5206\u6790\u5716\u50cf\u548c\u5716\u8868\u7684\u8996\u89ba\u80fd\u529b<\/li>\n<\/ul>\n<p><strong>Claude 3.5 Haiku<\/strong><\/p>\n<p>\u642d\u914dClaude 3 Opus \u6027\u80fd\u7684\u5168\u65b0\u9ad8\u6027\u50f9\u6bd4\u6a5f\u578b<\/p>\n<p>SWE-bench \u9a57\u8b49\u5f97\u5206\u70ba40.6%<\/p>\n<p>\u91dd\u5c0d\u9762\u5411\u5ba2\u6236\u7684\u61c9\u7528\u7a0b\u5f0f\u9032\u884c\u4e86\u6700\u4f73\u5316<\/p>\n<h3 id=\"heading-claude-35-sonnet-vs-chatgpt-4o-vs-gemini-15-pro\" class=\"permalink-heading\">Claude 3.5 Sonnet vs ChatGPT 4o vs Gemini 1.5 Pro<\/h3>\n<table style=\"border-collapse: collapse; width: 100%; height: 144px;\" border=\"1\">\n<tbody>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; text-align: center; height: 24px;\">\u80fd\u529b<\/td>\n<td style=\"width: 25%; text-align: center; height: 24px;\">Claude 3.5 Sonnet (New)<\/td>\n<td style=\"width: 25%; text-align: center; height: 24px;\">ChatGPT 4o<\/td>\n<td style=\"width: 25%; text-align: center; height: 24px;\">Gemini 1.5 Pro<\/td>\n<\/tr>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u591a\u6a21\u614b\u63a8\u7406\u5206\u6578<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">0.92<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">0.90<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">0.89<\/td>\n<\/tr>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; height: 24px; text-align: center;\">OCR\/\u624b\u5beb\u8b58\u5225<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u51fa\u8272\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u51fa\u8272\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u51fa\u8272\u7684<\/td>\n<\/tr>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u5716\u8868\/\u5716\u5f62\u89e3\u91cb<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u512a\u8d8a\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u597d\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u597d\u7684<\/td>\n<\/tr>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u8996\u89ba\u8cc7\u6599\u8655\u7406<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u5148\u9032\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u57fa\u672c\u7684<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u57fa\u672c\u7684<\/td>\n<\/tr>\n<tr style=\"height: 24px;\">\n<td style=\"width: 25%; height: 24px; text-align: center;\">\u4e0a\u4e0b\u6587\u8996\u7a97\u5927\u5c0f<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">20 \u842c\u500b\u4ee3\u5e63<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">8K \u4ee3\u5e63<\/td>\n<td style=\"width: 25%; height: 24px; text-align: center;\">8K \u4ee3\u5e63<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Claude 3.5 Sonnet \u5728\u591a\u6a21\u5f0f\u63a8\u7406\u4efb\u52d9\u4e2d\u8868\u73fe\u51fa\u8272\uff0c\u5c24\u5176\u64c5\u9577\uff1a<\/p>\n<ul>\n<li>\u8996\u89ba\u6578\u64da\u89e3\u91cb\u8207\u5206\u6790<\/li>\n<li>\u4f7f\u7528\u8996\u89ba\u5143\u7d20\u8655\u7406\u5927\u578b\u6587\u6a94<\/li>\n<li>\u9032\u968e\u5716\u8868\u548c\u5716\u5f62\u7406\u89e3<\/li>\n<\/ul>\n<p>\u9019\u4e09\u7a2e\u6a21\u578b\u5728OCR \u548c\u96e3\u4ee5\u8fa8\u8a8d\u7684\u624b\u5beb\u8b58\u5225\u7b49\u57fa\u672c\u8996\u89ba\u4efb\u52d9\u4e2d\u8868\u73fe\u540c\u6a23\u51fa\u8272\uff0c\u4f46Claude 3.5 Sonnet \u5728\u9700\u8981\u8a73\u7d30\u5206\u6790\u548c\u89e3\u91cb\u7684\u66f4\u8907\u96dc\u7684\u8996\u89ba\u63a8\u7406\u5834\u666f\u4e2d\u8868\u73fe\u51fa\u7279\u5225\u7684\u512a\u52e2\u3002<\/p>\n<h3 id=\"heading-claude-35-sonnet-a-mixed-bag-of-improvements-and-quirks\" class=\"permalink-heading\">Claude 3.5 Sonnet\uff1a\u6539\u9032\u8207\u7f3a\u9677\u4e26\u5b58<\/h3>\n<p>Claude 3.5 Sonnet \u7684\u6700\u65b0\u7248\u672c\u5728AI \u793e\u7fa4\u4e2d\u5f15\u8d77\u4e86\u6975\u5927\u7684\u8f5f\u52d5\uff0c\u7528\u6236\u5831\u544a\u4e86\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u6539\u9032\u548c\u610f\u60f3\u4e0d\u5230\u7684\u6311\u6230\u3002\u4ee5\u4e0b\u5168\u9762\u4ecb\u7d39\u958b\u767c\u4eba\u54e1\u548c\u4f7f\u7528\u8005\u5c0d\u65b0\u6a21\u578b\u7684\u9ad4\u9a57\u3002<\/p>\n<h3 id=\"heading-code-generation-and-development\" class=\"permalink-heading\">\u7a0b\u5f0f\u78bc\u751f\u6210\u548c\u958b\u767c<\/h3>\n<p><strong>iOS \u958b\u767c\u6210\u529f<\/strong>\u5e7e\u4f4d\u958b\u767c\u4eba\u54e1\u5831\u544a\u4e86\u4f7f\u7528Sonnet 3.5 \u958b\u767ciOS \u61c9\u7528\u7a0b\u5f0f\u7684\u6b63\u9762\u9ad4\u9a57\uff0c\u4e26\u6307\u51fa\u89e3\u6c7a\u554f\u984c\u7684\u80fd\u529b\u6709\u986f\u8457\u63d0\u9ad8[1]\u3002\u8a72\u6a21\u578b\u5c55\u793a\u4e86\u589e\u5f37\u7684\u89e3\u6c7a\u8907\u96dc\u7de8\u78bc\u554f\u984c\u7684\u80fd\u529b\uff0c\u5118\u7ba1\u4e00\u4e9b\u7528\u6236\u6307\u51fa\u5176\u6027\u80fd\u5b58\u5728\u4e0d\u4e00\u81f4\u3002<\/p>\n<p><strong>\u6574\u5408\u5de5\u4f5c\u6d41\u7a0b<\/strong>\u958b\u767c\u4eba\u54e1\u5df2\u7d93\u5c07Claude \u8207\u5404\u7a2e\u5de5\u5177\u7d50\u5408\u5efa\u7acb\u4e86\u6709\u6548\u7684\u5de5\u4f5c\u6d41\u7a0b\uff1a<\/p>\n<ul>\n<li>\u5e38\u898f\u67e5\u8a62\u7684Web \u4ecb\u9762<\/li>\n<li>\u900f\u904eBolt Mac \u61c9\u7528\u7a0b\u5f0f\u9032\u884cAPI \u96c6\u6210<\/li>\n<li>\u7528\u65bc\u76f4\u63a5\u7a0b\u5f0f\u78bc\u4e92\u52d5\u7684\u904a\u6a19<\/li>\n<li>\u7528\u65bc\u7ba1\u7406\u5c08\u6848\u6587\u4ef6\u7684\u81ea\u8a02Python \u8173\u672c<\/li>\n<\/ul>\n<h3 id=\"heading-notable-behavioral-changes\" class=\"permalink-heading\">\u986f\u8457\u7684\u884c\u70ba\u8b8a\u5316<\/h3>\n<p><strong>\u500b\u6027\u589e\u5f37<\/strong>\u4f7f\u7528\u8005\u767c\u73feSonnet 3.5 \u5728\u5c0d\u8a71\u4e2d\u8868\u73fe\u51fa\u66f4\u591a\u7684\u500b\u6027\u548c\u53c3\u8207\u5ea6\uff0c\u6709\u4eba\u6307\u51fa\u5b83\u5728\u4e92\u52d5\u4e2d\u300c\u8d85\u7d1a\u89aa\u5207\u300d\u548c\u300c\u4e0d\u53ef\u601d\u8b70\u300d[1]\u3002\u8207\u5148\u524d\u7684\u7248\u672c\u76f8\u6bd4\uff0c\u8a72\u6a21\u578b\u5728\u56de\u61c9\u4e2d\u8868\u73fe\u51fa\u66f4\u5927\u7684\u81ea\u4fe1\u548c\u667a\u6167\u3002<\/p>\n<p><strong>\u4e00\u81f4\u6027\u6311\u6230<\/strong>\u4e00\u4e9b\u7528\u6236\u5831\u544a\u4e86\u4e0d\u4e00\u81f4\u7684\u884c\u70ba\uff1a<\/p>\n<ul>\n<li>\u5076\u723e\u6703\u4e0d\u5fc5\u8981\u5730\u5206\u88c2\u56de\u61c9<\/li>\n<li>\u8655\u7406\u8907\u96dc\u67e5\u8a62\u6642\u7684\u6548\u80fd\u4e0d\u7a69\u5b9a<\/li>\n<li>\u6703\u8a71\u4e4b\u9593\u7684\u97ff\u61c9\u54c1\u8cea\u6ce2\u52d5<\/li>\n<\/ul>\n<h3 id=\"heading-technical-limitations\" class=\"permalink-heading\">\u6280\u8853\u9650\u5236<\/h3>\n<p><strong>\u901f\u7387\u9650\u5236<\/strong>\u4f7f\u7528\u8005\u5df2\u7d93\u6ce8\u610f\u5230\u901f\u7387\u9650\u5236\u7684\u6311\u6230\uff0c\u7279\u5225\u662f\u5728\u8655\u7406\u5927\u578b\u5c08\u6848\u6216\u9577\u6642\u9593\u5c0d\u8a71\u6642\u3002\u57fa\u65bc\u4ee4\u724c\u7684\u914d\u984d\u7cfb\u7d71\u9700\u8981\u5c0d\u5c0d\u8a71\u60c5\u5883\u9032\u884c\u7b56\u7565\u7ba1\u7406\uff0c\u4ee5\u6700\u5927\u9650\u5ea6\u5730\u63d0\u9ad8\u6548\u7387\u3002<\/p>\n<p><strong>\u4ee3\u78bc\u4fee\u6539\u554f\u984c<\/strong>\u4e00\u4e9b\u958b\u767c\u4eba\u54e1\u5831\u544a\u4e86\u7a0b\u5f0f\u78bc\u4fee\u6539\u65b9\u9762\u7684\u6311\u6230\uff1a<\/p>\n<ul>\n<li>\u7a0b\u5f0f\u78bc\u589e\u5f37\u904e\u7a0b\u4e2d\u5076\u723e\u6703\u522a\u9664\u91cd\u8981\u529f\u80fd<\/li>\n<li>\u5132\u5b58\u548c\u5feb\u53d6\u6307\u4ee4\u8655\u7406\u4e0d\u4e00\u81f4<\/li>\n<li>\u9700\u8981\u591a\u500b\u63d0\u793a\u624d\u80fd\u7dad\u6301\u6240\u9700\u7684\u529f\u80fd<\/li>\n<\/ul>\n<h3 id=\"heading-conclusion\" class=\"permalink-heading\">\u7d50\u8ad6<\/h3>\n<p>\u96d6\u7136Claude 3.5 Sonnet \u5728\u8a31\u591a\u9818\u57df\u90fd\u53d6\u5f97\u4e86\u91cd\u5927\u9032\u6b65\uff0c\u4f46\u5176\u6548\u80fd\u53d6\u6c7a\u65bc\u7279\u5b9a\u7528\u4f8b\u548c\u5be6\u4f5c\u65b9\u6cd5\u3002\u5efa\u8b70\u4f7f\u7528\u8005\u5236\u5b9a\u9069\u7576\u7684\u5de5\u4f5c\u6d41\u7a0b\u548c\u7b56\u7565\uff0c\u4ee5\u6700\u5927\u9650\u5ea6\u5730\u767c\u63ee\u5176\u512a\u52e2\uff0c\u540c\u6642\u514b\u670d\u5176\u9650\u5236\u3002<\/p>","protected":false},"excerpt":{"rendered":"<p>\u4e86\u89e3Claude Sonnet 3.5 \u548cClaude 3.5 Haiku \u7684\u6700\u65b0\u529f\u80fd\u548c\u589e\u5f37\u529f\u80fd\uff0c\u5305\u62ec\u6539\u9032\u7684\u6548\u80fd\u3001\u65b0\u529f\u80fd\u548c\u4f7f\u7528\u8005\u53cb\u597d\u7684\u66f4\u65b0\u3002\u96a8\u6642\u4e86\u89e3\u9019\u4e9b\u7248\u672c\u5728\u9ad8\u968eAI \u8a69\u6b4c\u5de5\u5177\u9818\u57df\u7684\u7368\u7279\u4e4b\u8655\u3002<\/p>","protected":false},"author":1,"featured_media":1229,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[65],"tags":[],"class_list":["post-1228","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technical-sharing"],"_links":{"self":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/1228","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/comments?post=1228"}],"version-history":[{"count":0,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/1228\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media\/1229"}],"wp:attachment":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media?parent=1228"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/categories?post=1228"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/tags?post=1228"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}