{"id":11377,"date":"2026-08-25T00:43:07","date_gmt":"2026-08-24T16:43:07","guid":{"rendered":"https:\/\/infernews.com\/blog\/agentmercury-ai\/"},"modified":"2026-08-25T00:44:52","modified_gmt":"2026-08-24T16:44:52","slug":"agentmercury-ai","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/agentmercury-ai\/","title":{"rendered":"AgentMercury \u751f\u6210\u53ef\u9a57\u8b49\u5546\u696d\u4e16\u754c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-562b21d13c3b.jpg\" alt=\"Og image\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576AI\u4ee3\u7406\u8981\u8655\u7406\u8de8\u90e8\u9580\u4f01\u696d\u6d41\u7a0b\uff0c\u96e3\u9ede\u5f80\u5f80\u4e0d\u53ea\u5728\u65bc\u9078\u64c7\u5de5\u5177\uff0c\u9084\u8981\u5728\u591a\u500b\u670d\u52d9\u4e4b\u9593\u7dad\u6301\u4e00\u81f4\u72c0\u614b\uff0c\u6700\u5f8c\u4ea4\u51fa\u53ef\u4ee5\u6838\u5be6\u7684\u7d50\u679c\u3002AgentMercury\u662f\u4e00\u500b\u7528\u65bc\u8a13\u7df4 Computer-use agents\uff08CUAs\uff09\u7684\u74b0\u5883\u751f\u6210\u9805\u76ee\uff0c\u8b93\u9ad8\u5c64\u6b21\u5546\u696d\u60c5\u5883\u76f4\u63a5\u8b8a\u6210\u53ef\u57f7\u884c\u3001\u53ef\u91cd\u64ad\u548c\u53ef\u9a57\u8b49\u7684\u865b\u64ec\u4e16\u754c\uff0c\u4ee3\u7406\u53ef\u4ee5\u5728\u5176\u4e2d\u9032\u884c\u591a\u8f2a\u64cd\u4f5c\uff0c\u800c\u7cfb\u7d71\u6703\u6309\u5b8c\u6210\u5f8c\u7684\u72c0\u614b\u8a08\u7b97\u734e\u52f5\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5e38\u898b\u505a\u6cd5\u6703\u5148\u70ba\u6307\u5b9a\u4efb\u52d9\u6216 benchmark \u624b\u5de5\u5efa\u7acb\u74b0\u5883\uff0c\u74b0\u5883\u81ea\u7136\u53ea\u8986\u84cb\u6e2c\u8a66\u8005\u9810\u5148\u60f3\u5230\u7684\u6d41\u7a0b\u3002AgentMercury\u5247\u628a\u74b0\u5883\u5efa\u69cb\u672c\u8eab\u4ea4\u7d66 Planet\uff1a\u5b83\u7531\u516c\u53f8\u8eab\u4efd\u3001\u670d\u52d9\u5716\u3001\u8cc7\u6599\u8868\u7d50\u69cb\u3001\u521d\u59cb\u72c0\u614b\u548c\u8de8\u670d\u52d9\u4e0d\u8b8a\u689d\u4ef6\u7d44\u6210\u4e00\u500b\u5b8c\u6574\u4e16\u754c\uff0c\u518d\u7531 Task \u5728\u5171\u4eab\u4e16\u754c\u4e0a\u52a0\u5165\u4e0d\u540c\u76ee\u6a19\u548c\u8a55\u5206\u898f\u5247\u3002\u4e00\u500b\u4e16\u754c\u53ef\u4ee5\u627f\u8f09\u591a\u500b\u4efb\u52d9\uff0c\u6e1b\u5c11\u6bcf\u500b\u65b0\u6d41\u7a0b\u90fd\u8981\u91cd\u65b0\u9020\u74b0\u5883\u7684\u6210\u672c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9a57\u8b49\u689d\u4ef6\u4e3b\u8981\u4ee5\u74b0\u5883\u8cc7\u6599\u5eab\u4e0a\u7684 SQL \u5f62\u5f0f\u57f7\u884c\u3002\u4ee3\u7406\u5b8c\u6210\u4e00\u8f2a\u64cd\u4f5c\u5f8c\uff0c\u7cfb\u7d71\u6703\u6839\u64da\u4efb\u52d9 rubric \u548c\u96b1\u85cf\u7684\u4e0d\u8b8a\u689d\u4ef6\uff0c\u5c0d\u6574\u689d trajectory \u9032\u884c\u78ba\u5b9a\u6027\u8a55\u5206\uff1b\u74b0\u5883\u4e0d\u6703\u5728\u6bcf\u4e00\u6b65\u66ff\u4ee3\u7406\u5224\u65b7\u5c0d\u932f\uff0c\u4ee3\u7406\u5fc5\u9808\u81ea\u884c\u8655\u7406\u8de8\u670d\u52d9\u8981\u6c42\uff0c\u5b8c\u6210\u5f8c\u624d\u63a5\u53d7\u6aa2\u67e5\u3002\u9019\u7a2e\u5b89\u6392\u8b93\u7d50\u679c\u53ef\u4ee5\u91cd\u64ad\uff0c\u4e5f\u4ee4 Reinforcement Learning\uff08RL\uff09\u8a13\u7df4\u5f97\u5230\u8f03\u7a69\u5b9a\u7684\u56de\u994b\u3002<\/p>\n\n\n<div\n    \tclass=\"align wp-block-vpb-video\"    id='vpbpVideoPlayer-1'\n    data-attributes='{&quot;source&quot;:&quot;https:\\\/\\\/minstar.github.io\\\/AgentMercury\\\/assets\\\/agentmercury_teaser.mp4&quot;,&quot;repeat&quot;:true,&quot;autoplay&quot;:true,&quot;muted&quot;:true,&quot;align&quot;:&quot;&quot;,&quot;poster&quot;:&quot;&quot;,&quot;controls&quot;:{&quot;play-large&quot;:true,&quot;restart&quot;:false,&quot;rewind&quot;:true,&quot;play&quot;:true,&quot;fast-forward&quot;:true,&quot;progress&quot;:true,&quot;current-time&quot;:true,&quot;duration&quot;:false,&quot;mute&quot;:true,&quot;volume&quot;:true,&quot;pip&quot;:false,&quot;airplay&quot;:false,&quot;captions&quot;:false,&quot;settings&quot;:true,&quot;download&quot;:false,&quot;fullscreen&quot;:true},&quot;captions&quot;:[],&quot;ratio&quot;:&quot;16:9&quot;,&quot;accentColor&quot;:&quot;&quot;,&quot;lazyLoad&quot;:false,&quot;width&quot;:&quot;100%&quot;,&quot;radius&quot;:&quot;0px&quot;,&quot;resetOnEnd&quot;:false,&quot;autoHideControl&quot;:true,&quot;isSetup&quot;:false}'\n><\/div>\n\n\n<p class=\"wp-block-paragraph\">\u76ee\u524d\u9805\u76ee\u6574\u7406\u51fa4,783\u500b executable worlds\uff0c\u6db5\u84cb14\u500b\u7522\u696d\u548c50\u500b\u570b\u5bb6\uff0c\u5305\u542b13.98M seeded state rows\u3001842\u7a2e\u5de5\u5177\u548c43,300\u500b\u53ef\u9a57\u8b49 RL \u4efb\u52d9\u3002\u5178\u578b\u4e16\u754c\u7d04\u670913\u9805\u670d\u52d9\u300165\u7a2e\u5de5\u5177\u300131\u5f35\u72c0\u614b\u8868\uff0c\u4ee5\u53ca\u7d0415\u500b\u4ee5 SQL \u7de8\u5beb\u7684 deterministic verifiers\uff1b\u8a13\u7df4\u8cc7\u6599\u6d89\u53ca4,326\u500b\u516c\u53f8\u74b0\u5883\uff0c\u4f46 Qwen3.5-4B \u7684\u5831\u544a\u53ea\u4f7f\u7528\u5176\u4e2d3,200\u500b\u4efb\u52d9\u53d6\u5f97\u68af\u5ea6\uff0c\u4ecd\u8986\u84cb53.5%\u7684\u74b0\u5883\u300162.9%\u7684\u7522\u696d\u548c75.8%\u7684\u5de5\u5177\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u4ee5 Qwen3.5-4B \u548c Qwen3.5-35B-A3B \u9032\u884c RL \u8a13\u7df4\uff0c\u4ee3\u7406\u900f\u904e Model Context Protocol\uff08MCP\uff09\u64cd\u4f5c\u5373\u6642\u5f62\u5f0f\u7684\u5546\u696d\u8edf\u4ef6\uff0c\u4e26\u4ee5 GRPO\uff08Group Relative Policy Optimization\uff09\u70ba\u4e3b\u8981\u512a\u5316\u65b9\u6cd5\uff0cSAO \u5247\u4f5c\u70ba\u53e6\u4e00\u7a2e\u9078\u9805\u3002Qwen3.5-4B \u7684\u5e73\u5747\u734e\u52f5\u7d04\u75310.25\u5347\u81f30.53\uff0c\u622a\u65b7\u56de\u61c9\u6bd4\u4f8b\u753178%\u964d\u81f33%\uff1b\u5728\u672a\u53c3\u8207\u8a13\u7df4\u7684 EnterpriseOps-Gym \u516b\u500b\u4f01\u696d\u9818\u57df\uff0c\u5e73\u5747\u5206\u6578\u753112.3\u5347\u81f315.7\uff0c\u589e\u5e45\u70ba27.6%\u300235B-A3B \u4f7f\u7528 GRPO \u548c SAO \u5f8c\uff0c\u5e73\u5747\u5206\u6578\u5206\u5225\u753124.8\u5347\u81f328.1\u53ca28.3\uff0c\u4f46\u7d50\u679c\u4ecd\u5c6c\u9805\u76ee\u6240\u5831\u544a\u7684\u8a13\u7df4\u548c\u8a55\u6e2c\u7bc4\u570d\uff0c\u672a\u4ee3\u8868\u4ee3\u7406\u5df2\u80fd\u53ef\u9760\u8655\u7406\u6240\u6709\u4f01\u696d\u5de5\u4f5c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5c07\u9ad8\u5c64\u6b21\u5546\u696d\u60c5\u5883\u7de8\u8b6f\u6210\u53ef\u57f7\u884c\u4e16\u754c\uff0c\u800c\u4e0d\u662f\u53ea\u70ba\u55ae\u4e00\u4efb\u52d9\u9020\u74b0\u5883<\/li>\n\n\n\n<li>\u4ee5\u5171\u4eab\u8cc7\u6599\u5eab\u3001SQL \u9a57\u8b49\u689d\u4ef6\u548c\u53ef\u91cd\u64ad\u8a55\u5206\u6aa2\u67e5\u8de8\u670d\u52d9\u7d50\u679c<\/li>\n\n\n\n<li>\u63d0\u4f9b4,783\u500b\u74b0\u5883\u3001842\u7a2e\u5de5\u5177\u548c43,300\u500b\u53ef\u9a57\u8b49 RL \u4efb\u52d9<\/li>\n\n\n\n<li>Qwen3.5 \u7cfb\u5217\u5728\u591a\u8f2a MCP \u5de5\u5177\u64cd\u4f5c\u8a13\u7df4\u5f8c\uff0c\u734e\u52f5\u548c\u672a\u898b\u9818\u57df\u8868\u73fe\u5747\u6709\u6539\u5584<\/li>\n\n\n\n<li>\u4ecd\u9700\u7559\u610f\u5408\u6210\u74b0\u5883\u8207\u771f\u5be6\u4f01\u696d\u8edf\u4ef6\u4e4b\u9593\u7684\u5dee\u8ddd\uff0c\u4ee5\u53ca\u8a55\u6e2c\u7bc4\u570d\u6709\u9650\u7684\u554f\u984c<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/minstar.github.io\/AgentMercury\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/collections\/Minbyul\/agentmercury\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u628a\u4f01\u696d\u60c5\u5883\u8f49\u6210\u53ef\u57f7\u884c\u74b0\u5883\uff0c\u8b93AI\u4ee3\u7406\u5728\u591a\u670d\u52d9\u6d41\u7a0b\u4e2d\u8a13\u7df4\uff0c\u518d\u8f49\u79fb\u5230\u672a\u898b\u904e\u7684\u8a55\u6e2c\u9805\u76ee\u3002<\/p>\n","protected":false},"author":8,"featured_media":11376,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,185,116,151,127,199],"tags":[],"class_list":["post-11377","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-agentic","category-mcp","category-127","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11377","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=11377"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11377\/revisions"}],"predecessor-version":[{"id":11380,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11377\/revisions\/11380"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11376"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11377"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11377"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11377"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}