
{"id":13513,"date":"2026-10-17T02:56:55","date_gmt":"2026-10-16T18:56:55","guid":{"rendered":"https:\/\/infernews.com\/blog\/opengcua-2\/"},"modified":"2026-10-17T03:05:20","modified_gmt":"2026-10-16T19:05:20","slug":"opengcua-2","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/opengcua-2\/","title":{"rendered":"OpenGCUA \u8b93\u96fb\u8166\u4ee3\u7406\u8de8\u8d8a GUI \u9650\u5236"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/OpenGCUA-GUI-1.jpg\" alt=\"OpenGCUA. Open the full stack of generalist computer use. GPT-6 made computer use the frontier. We make it open. Tested\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u96fb\u8166\u4ee3\u7406\u4e0d\u518d\u53ea\u9760\u6ed1\u9f20\u9ede\u64ca\uff0c\u8655\u7406\u8907\u96dc\u5de5\u4f5c\u4fbf\u53ef\u4ee5\u540c\u6642\u8b80\u53d6\u756b\u9762\u3001Accessibility Tree\u3001\u61c9\u7528\u7a0b\u5f0f API \u548c\u547d\u4ee4\u5217\u3002OpenGCUA \u662f\u4e00\u500b\u958b\u6e90 Computer-use agents\uff08CUAs\uff09\u9805\u76ee\uff0c\u5efa\u69cb generalist computer-use agent\uff08GCUA\uff09\u5b8c\u6574\u5806\u758a\uff0c\u89e3\u6c7a GUI-only \u4ee3\u7406\u5728\u6548\u7387\u548c\u64cd\u4f5c\u5f48\u6027\u4e0a\u7684\u9650\u5236\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u6703\u6309\u4efb\u52d9\u6c7a\u5b9a\u8981\u89c0\u5bdf\u751a\u9ebc\u3001\u63a1\u53d6\u54ea\u7a2e\u884c\u52d5\uff0c\u4ee5\u53ca\u5982\u4f55\u6574\u7406\u4e0a\u4e0b\u6587\uff1b\u9664\u4e86 GUI \u64cd\u4f5c\uff0c\u4e5f\u80fd\u547c\u53eb app API\u3001\u57f7\u884c shell commands\uff0c\u4e26\u900f\u904e code execution \u628a<a href=\"https:\/\/tongyi-mai.github.io\/OpenGCUA\/#lab\" target=\"_blank\" rel=\"noopener\"><mark>\u591a\u500b\u6b65\u9a5f\u7d44\u6210<\/mark><\/a>\u5e36\u6709\u8ff4\u5708\u548c\u689d\u4ef6\u7684\u7a0b\u5f0f\u3002OSWorld \u6574\u5408\u7bc4\u4f8b\u63d0\u4f9b screenshot-action \u548c code-action \u8def\u5f91\uff0cmacOS 14+ harness \u5247\u4ee5\u9078\u55ae\u5217\u61c9\u7528\u7a0b\u5f0f\u5728\u672c\u6a5f\u88dd\u7f6e\u57f7\u884c\u4ee3\u7406\uff0c\u4ecd\u5c6c preliminary\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">OpenGCUA \u4ea6\u63d0\u4f9b Cross-Device SDK\uff0c\u5617\u8a66\u7528\u540c\u4e00\u5957 programming interface \u9023\u63a5\u96fb\u8166\u3001\u700f\u89bd\u5668\u548c\u624b\u6a5f\u3002\u9019\u6bd4\u56fa\u5b9a\u898f\u5247\u7684\u5de5\u4f5c\u6d41\u66f4\u9748\u6d3b\uff0c\u4f46\u4e5f\u4ee3\u8868\u547d\u4ee4\u3001API \u548c GUI \u6b0a\u9650\u9700\u8981\u66f4\u56b4\u683c\u7684\u9694\u96e2\uff1b\u9805\u76ee\u76ee\u524d\u4ecd\u5728\u9010\u6b65\u958b\u653e\u6a21\u578b\u3001harness\u3001SDK\u3001trajectories\u3001\u6548\u7387\u512a\u5316\u548c\u8a55\u4f30\u5de5\u5177\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">OSWorld-v2 \u7684 82 \u500b offline tasks \u4e2d\uff0c\u516b\u500b frontier models \u7684\u5e73\u5747 partial score \u7531 GUI-only execution \u7684 45.3 \u63d0\u5347\u81f3 64.9\uff0c\u589e\u52a0 19.6 \u5206\uff1b\u6bcf\u9805\u4efb\u52d9\u5e73\u5747\u6b65\u6578\u6e1b\u5c11 53%\uff0c\u516d\u500b\u6a21\u578b\u7684 output tokens \u4ea6\u6e1b\u5c11 27.2%\u3002\u8f03\u5f31\u6a21\u578b\u7372\u76ca\u8f03\u5927\uff0c\u4f8b\u5982 GLM-5.3-Flash \u589e\u52a0 35.2 \u5206\uff0cDeepSeek-V4.1-Flash \u589e\u52a0 29.8 \u5206\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u540c\u6642\u652f\u63f4 GUI\u3001AX Tree\u3001CLI\u3001API \u548c code execution<\/li>\n\n\n\n<li>\u9644\u6709\u516b\u500b\u6a21\u578b\u7684 evaluation trajectories\uff0c\u5305\u62ec GPT-6 Astra\u3001Claude Opus 5 \u548c Gemini 3.8 Flash<\/li>\n\n\n\n<li>\u53ef\u7528 OSWorld \u6574\u5408\u7bc4\u4f8b\u7814\u7a76 screenshot-action \u8207 code-action<\/li>\n\n\n\n<li>macOS harness \u548c Cross-Device SDK \u65b9\u4fbf\u7814\u7a76\u672c\u6a5f\u53ca\u8de8\u88dd\u7f6e\u5de5\u4f5c\u6d41<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u7814\u7a76 Computer-use agents\u3001\u9700\u8981\u5206\u6790\u4ee3\u7406\u884c\u70ba\uff0c\u6216\u5e0c\u671b\u628a\u96fb\u8166\u3001\u700f\u89bd\u5668\u548c\u624b\u6a5f\u64cd\u4f5c\u7d0d\u5165\u540c\u4e00\u5de5\u4f5c\u6d41\u7684\u5718\u968a\uff0c\u6703\u8f03\u5bb9\u6613\u5f9e\u4e2d\u53d7\u76ca\u3002\u9805\u76ee\u9069\u5408\u7528\u4f5c\u7814\u7a76\u548c\u5be6\u9a57\u57fa\u790e\uff0c\u672a\u5b9c\u76f4\u63a5\u8996\u70ba\u6210\u719f\u7684\u65e5\u5e38\u81ea\u52d5\u5316\u7522\u54c1\uff1b\u672c\u6a5f\u57f7\u884c\u3001\u6b0a\u9650\u63a7\u5236\u3001\u6a21\u578b\u76f8\u5bb9\u6027\u53ca\u4ecd\u5728\u958b\u767c\u7684\u6a21\u7d44\uff0c\u90fd\u6703\u5f71\u97ff\u90e8\u7f72\u96e3\u5ea6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/tongyi-mai.github.io\/OpenGCUA\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/Tongyi-MAI\/OpenGCUA\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>OpenGCUA \u628a\u8996\u89ba\u64cd\u4f5c\u3001AX Tree\u3001API \u8207\u6307\u4ee4\u6574\u5408\uff0c\u8b93\u4ee3\u7406\u6309\u4efb\u52d9\u9078\u64c7\u5b8c\u6210\u5de5\u4f5c\u7684\u65b9\u5f0f\u3002<\/p>\n","protected":false},"author":8,"featured_media":13512,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,171,116,76,185,147,140,137,38,166,146,191,199],"tags":[],"class_list":["post-13513","post","type-post","status-publish","format-standard","hentry","category-133","category-171","category-agentic","category-76","category-qwen","category-deepseek","category-gemini","category-api","category-38","category-mac","category-146","category-anthropic","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13513","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=13513"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13513\/revisions"}],"predecessor-version":[{"id":13515,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13513\/revisions\/13515"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/13512"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=13513"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=13513"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=13513"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}