{"id":92234,"date":"2026-08-09T19:56:41","date_gmt":"2026-08-09T11:56:41","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/92234.html"},"modified":"2026-08-09T19:56:41","modified_gmt":"2026-08-09T11:56:41","slug":"%e8%bf%91%e6%9c%9fai%e7%83%ad%e7%82%b9007%ef%bd%9carc-agi-3-%e5%be%97%e5%88%86%e4%b8%ba%e4%bd%95%e6%8f%90%e9%ab%98%e5%88%b0%e8%bf%91%e4%b8%89%e5%80%8d%ef%bc%9a%e6%a8%a1%e5%9e%8b%e4%b9%8b%e5%a4%96","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/92234.html","title":{"rendered":"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf"},"content":{"rendered":"<h2>\u8fd1\u671fAI\u70ed\u70b9007&#xff5c;ARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d&#xff1a;\u6a21\u578b\u4e4b\u5916&#xff0c;Agent Harness \u4e5f\u662f\u53d8\u91cf<\/h2>\n<p>\u4e3b\u8981\u4fe1\u606f\u6e90&#xff1a;OpenAI ARC-AGI-3 \u5b98\u65b9\u5b9e\u9a8c\u6587\u7ae0\u3001Responses API \u5b98\u65b9\u4ecb\u7ecd\u3001Compaction \u5b98\u65b9\u6307\u5357\u3001ARC Prize \u65b9\u6cd5\u8bf4\u660e<br \/>\n\u5173\u952e\u8bcd&#xff1a;ARC-AGI-3\u3001Agent Harness\u3001Responses API\u3001Compaction\u3001\u4e0a\u4e0b\u6587\u7ba1\u7406\u3001AI Agent\u3001\u6a21\u578b\u8bc4\u6d4b<\/p>\n<p>OpenAI \u5b9e\u9a8c\u539f\u6587&#xff1a;https:\/\/openai.com\/index\/how-two-settings-tripled-our-arc-agi-3-scores\/<br \/>\nResponses API&#xff1a;https:\/\/developers.openai.com\/blog\/responses-api<br \/>\nCompaction \u6307\u5357&#xff1a;https:\/\/developers.openai.com\/api\/docs\/guides\/compaction<\/p>\n<p>2026 \u5e74 7 \u6708 29 \u65e5&#xff0c;OpenAI \u516c\u5e03\u4e86\u4e00\u7ec4\u5bb9\u6613\u88ab\u8bef\u8bfb\u7684\u6570\u636e&#xff1a;\u5728 ARC-AGI-3 \u516c\u5f00\u4efb\u52a1\u96c6\u4e0a&#xff0c;GPT-5.6 Sol \u4f7f\u7528\u901a\u7528 Harness \u65f6\u5f97\u5206\u4e3a 13.3%&#xff1b;\u4fdd\u7559\u63a8\u7406\u72b6\u6001\u5e76\u542f\u7528\u4e0a\u4e0b\u6587\u538b\u7f29\u540e&#xff0c;\u5f97\u5206\u63d0\u9ad8\u5230 38.3%&#xff0c;\u540c\u65f6\u8f93\u51fa Token \u7ea6\u964d\u81f3\u539f\u6765\u7684\u516d\u5206\u4e4b\u4e00\u3002<\/p>\n<p>\u6a21\u578b\u6ca1\u6709\u66f4\u6362&#xff0c;\u6743\u91cd\u6ca1\u6709\u66f4\u65b0&#xff0c;\u4e3b\u8981\u53d8\u5316\u53d1\u751f\u5728\u6a21\u578b\u5916\u90e8\u7684\u8fd0\u884c\u7cfb\u7edf\u3002<\/p>\n<p>\u8fd9\u9879\u5b9e\u9a8c\u771f\u6b63\u503c\u5f97\u5f00\u53d1\u8005\u5173\u6ce8\u7684&#xff0c;\u4e0d\u662f\u201c\u67d0\u4e2a\u6a21\u578b\u53c8\u5237\u65b0\u4e86\u699c\u5355\u201d&#xff0c;\u800c\u662f\u4e00\u4e2a\u66f4\u63a5\u8fd1\u5de5\u7a0b\u73b0\u5b9e\u7684\u95ee\u9898&#xff1a;\u6211\u4eec\u901a\u5e38\u8bf4\u81ea\u5df1\u5728\u8bc4\u6d4b\u6a21\u578b&#xff0c;\u5b9e\u9645\u8bc4\u6d4b\u5230\u7684\u5f80\u5f80\u662f\u6a21\u578b\u3001API \u8bbe\u7f6e\u3001\u4e0a\u4e0b\u6587\u7ba1\u7406\u3001\u5de5\u5177\u63a5\u53e3\u548c\u4efb\u52a1\u5faa\u73af\u7684\u7ec4\u5408\u3002<\/p>\n<h3>ARC-AGI-3 \u6d4b\u8bd5\u7684\u662f\u4ec0\u4e48<\/h3>\n<p>ARC-AGI-3 \u4f7f\u7528\u964c\u751f\u7684\u4e8c\u7ef4\u6e38\u620f\u6d4b\u8bd5 Agent \u7684\u5b66\u4e60\u548c\u63a8\u7406\u80fd\u529b\u3002\u6a21\u578b\u4e0d\u4f1a\u83b7\u5f97\u5b8c\u6574\u89c4\u5219&#xff0c;\u9700\u8981\u6839\u636e\u8fde\u7eed\u753b\u9762\u548c\u64cd\u4f5c\u53cd\u9988\u63a8\u65ad&#xff1a;<\/p>\n<ul>\n<li>\u54ea\u4e9b\u5bf9\u8c61\u53ef\u4ee5\u4ea4\u4e92&#xff1b;<\/li>\n<li>\u4e00\u4e2a\u52a8\u4f5c\u4f1a\u600e\u6837\u6539\u53d8\u73af\u5883&#xff1b;<\/li>\n<li>\u5f53\u524d\u5173\u5361\u7684\u76ee\u6807\u662f\u4ec0\u4e48&#xff1b;<\/li>\n<li>\u8fc7\u53bb\u5931\u8d25\u7684\u5c1d\u8bd5\u8bf4\u660e\u4e86\u4ec0\u4e48&#xff1b;<\/li>\n<li>\u4e0b\u4e00\u6b65\u5e94\u8be5\u63a2\u7d22\u8fd8\u662f\u6267\u884c\u8ba1\u5212\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u7c7b\u4efb\u52a1\u548c\u9759\u6001\u95ee\u7b54\u4e0d\u540c\u3002\u6a21\u578b\u4e0d\u662f\u8bfb\u5b8c\u4e00\u9053\u9898\u540e\u4e00\u6b21\u8f93\u51fa\u7b54\u6848&#xff0c;\u800c\u662f\u5728\u4e00\u4e2a\u5faa\u73af\u91cc\u4e0d\u65ad\u884c\u52a8&#xff1a;<\/p>\n<p>\u89c2\u5bdf\u73af\u5883<br \/>\n   \u2193<br \/>\n\u5f62\u6210\u5047\u8bbe\u4e0e\u8ba1\u5212<br \/>\n   \u2193<br \/>\n\u6267\u884c\u4e00\u4e2a\u52a8\u4f5c<br \/>\n   \u2193<br \/>\n\u8bfb\u53d6\u65b0\u7684\u73af\u5883\u72b6\u6001<br \/>\n   \u2193<br \/>\n\u4fee\u6b63\u5047\u8bbe\u5e76\u7ee7\u7eed<\/p>\n<p>\u5982\u679c\u6bcf\u8f6e\u7ed3\u675f\u540e\u90fd\u4e22\u6389\u4e0a\u4e00\u6b21\u63a8\u7406&#xff0c;Agent \u5c31\u9700\u8981\u53cd\u590d\u91cd\u65b0\u7406\u89e3\u89c4\u5219&#xff1b;\u5982\u679c\u4e0a\u4e0b\u6587\u8fc7\u957f\u65f6\u76f4\u63a5\u5220\u9664\u6700\u65e9\u8bb0\u5f55&#xff0c;\u5b83\u8fd8\u4f1a\u9010\u6e10\u5fd8\u8bb0\u5df2\u7ecf\u9a8c\u8bc1\u8fc7\u7684\u7ed3\u8bba\u3002<\/p>\n<h3>\u4e24\u5957 Harness \u7684\u5dee\u5f02<\/h3>\n<p>OpenAI \u5728\u6587\u7ae0\u4e2d\u6307\u51fa&#xff0c;\u6700\u521d\u4f7f\u7528\u7684\u901a\u7528 Harness \u5b58\u5728\u4e24\u4e2a\u5173\u952e\u884c\u4e3a\u3002<\/p>\n<p>\u7b2c\u4e00&#xff0c;\u6bcf\u6b21\u6e38\u620f\u64cd\u4f5c\u540e&#xff0c;\u6a21\u578b\u7684\u79c1\u6709\u63a8\u7406\u72b6\u6001\u90fd\u4f1a\u88ab\u4e22\u5f03\u3002\u4e0b\u4e00\u8f6e\u867d\u7136\u8fd8\u80fd\u770b\u5230\u90e8\u5206\u64cd\u4f5c\u8bb0\u5f55&#xff0c;\u5374\u770b\u4e0d\u5230\u4ea7\u751f\u8fd9\u4e9b\u64cd\u4f5c\u7684\u8ba1\u5212\u548c\u4e2d\u95f4\u5224\u65ad\u3002<\/p>\n<p>\u7b2c\u4e8c&#xff0c;Harness \u4f7f\u7528\u6eda\u52a8\u622a\u65ad\u63a7\u5236\u4e0a\u4e0b\u6587\u3002\u8bb0\u5f55\u8d85\u8fc7\u9650\u5236\u540e&#xff0c;\u6700\u65e9\u7684\u89c2\u5bdf\u548c\u52a8\u4f5c\u4f1a\u88ab\u76f4\u63a5\u79fb\u9664\u3002<\/p>\n<p>OpenAI \u968f\u540e\u7528\u66f4\u63a5\u8fd1 ChatGPT \u4e0e Codex \u751f\u4ea7\u73af\u5883\u7684\u65b9\u5f0f\u91cd\u65b0\u5b9e\u73b0 Harness&#xff1a;<\/p>\n<table>\n<tr>\u7ef4\u5ea6\u901a\u7528 Harness\u8c03\u6574\u540e\u7684 Harness<\/tr>\n<tbody>\n<tr>\n<td>\u63a8\u7406\u8fde\u7eed\u6027<\/td>\n<td>\u6bcf\u8f6e\u4e22\u5f03\u63a8\u7406\u72b6\u6001<\/td>\n<td>\u901a\u8fc7 Responses API \u5ef6\u7eed\u4e0a\u4e00\u8f6e\u54cd\u5e94<\/td>\n<\/tr>\n<tr>\n<td>\u957f\u4e0a\u4e0b\u6587\u5904\u7406<\/td>\n<td>\u5220\u9664\u6700\u65e9\u8bb0\u5f55<\/td>\n<td>\u4f7f\u7528 Compaction \u538b\u7f29\u5386\u53f2\u4fe1\u606f<\/td>\n<\/tr>\n<tr>\n<td>\u516c\u5f00\u96c6\u5f97\u5206<\/td>\n<td>13.3%<\/td>\n<td>38.3%<\/td>\n<\/tr>\n<tr>\n<td>\u8f93\u51fa Token<\/td>\n<td>\u57fa\u51c6\u503c<\/td>\n<td>\u7ea6\u4e3a\u539f\u6765\u7684\u516d\u5206\u4e4b\u4e00<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u5b98\u65b9\u6839\u636e\u516c\u5f00\u6e38\u620f\u8bb0\u5f55\u4f30\u8ba1&#xff0c;\u4eba\u7c7b\u6d4b\u8bd5\u8005\u5728\u540c\u4e00\u6307\u6807\u4e0a\u7684\u5e73\u5747\u5f97\u5206\u7ea6\u4e3a 48%\u3002\u56e0\u6b64&#xff0c;38.3% \u4ecd\u4e0d\u4ee3\u8868 Agent \u5df2\u7ecf\u8fbe\u5230\u7a33\u5b9a\u7684\u4eba\u7c7b\u6c34\u5e73&#xff0c;\u4f46\u8db3\u4ee5\u8bf4\u660e Harness \u4f1a\u663e\u8457\u5f71\u54cd\u7ed3\u679c\u3002<\/p>\n<h3>\u4fdd\u7559\u63a8\u7406\u72b6\u6001\u4e3a\u4ec0\u4e48\u6709\u7528<\/h3>\n<p>\u957f\u7a0b Agent \u7684\u96be\u70b9\u4e0d\u662f\u6bcf\u4e00\u6b65\u90fd\u9700\u8981\u91cd\u65b0\u505a\u4e00\u9053\u96be\u9898&#xff0c;\u800c\u662f\u9700\u8981\u5728\u51e0\u5341\u751a\u81f3\u51e0\u767e\u8f6e\u64cd\u4f5c\u4e2d\u4fdd\u6301\u540c\u4e00\u4e2a\u8ba1\u5212\u3002<\/p>\n<p>\u5047\u8bbe\u6a21\u578b\u5df2\u7ecf\u901a\u8fc7\u524d\u4e94\u8f6e\u64cd\u4f5c\u53d1\u73b0&#xff1a;<\/p>\n<p>\u84dd\u8272\u65b9\u5757\u53ef\u4ee5\u63a8\u52a8&#xff1b;<br \/>\n\u7ea2\u8272\u533a\u57df\u4f1a\u91cd\u7f6e\u5173\u5361&#xff1b;<br \/>\n\u53f3\u4e0a\u89d2\u7684\u95e8\u53ea\u6709\u5728\u4e24\u4e2a\u5f00\u5173\u540c\u65f6\u6fc0\u6d3b\u540e\u624d\u4f1a\u6253\u5f00\u3002<\/p>\n<p>\u5982\u679c\u7b2c\u516d\u8f6e\u5f00\u59cb\u65f6\u8fd9\u4e9b\u5224\u65ad\u4e0d\u518d\u5b58\u5728&#xff0c;\u6a21\u578b\u53ea\u80fd\u4ece\u52a8\u4f5c\u65e5\u5fd7\u4e2d\u91cd\u65b0\u63a8\u5bfc\u3002\u91cd\u590d\u63a8\u5bfc\u4f1a\u589e\u52a0\u8f93\u51fa Token&#xff0c;\u4e5f\u66f4\u5bb9\u6613\u8ba9\u7b56\u7565\u53d1\u751f\u6f02\u79fb\u3002<\/p>\n<p>Responses API \u652f\u6301\u901a\u8fc7\u4e0a\u4e00\u8f6e\u54cd\u5e94 ID \u5ef6\u7eed\u4efb\u52a1\u72b6\u6001\u3002\u4e0b\u9762\u662f\u4e00\u4e2a\u7cbe\u7b80\u7684 Agent \u5faa\u73af\u793a\u4f8b&#xff0c;\u91cd\u70b9\u5c55\u793a previous_response_id \u7684\u4f7f\u7528\u65b9\u5f0f&#xff1a;<\/p>\n<p><span class=\"token keyword\">import<\/span> json<br \/>\n<span class=\"token keyword\">import<\/span> os<\/p>\n<p><span class=\"token keyword\">from<\/span> openai <span class=\"token keyword\">import<\/span> OpenAI<\/p>\n<p>client <span class=\"token operator\">&#061;<\/span> OpenAI<span class=\"token punctuation\">(<\/span>api_key<span class=\"token operator\">&#061;<\/span>os<span class=\"token punctuation\">.<\/span>environ<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;OPENAI_API_KEY&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><br \/>\nmodel <span class=\"token operator\">&#061;<\/span> <span class=\"token string\">&#034;gpt-5.6-sol&#034;<\/span><\/p>\n<p><span class=\"token keyword\">def<\/span> <span class=\"token function\">build_observation<\/span><span class=\"token punctuation\">(<\/span>step<span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">,<\/span> frame<span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">dict<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&#8211;<\/span><span class=\"token operator\">&gt;<\/span> <span class=\"token builtin\">str<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token keyword\">return<\/span> json<span class=\"token punctuation\">.<\/span>dumps<span class=\"token punctuation\">(<\/span><br \/>\n        <span class=\"token punctuation\">{<\/span><br \/>\n            <span class=\"token string\">&#034;step&#034;<\/span><span class=\"token punctuation\">:<\/span> step<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;observation&#034;<\/span><span class=\"token punctuation\">:<\/span> frame<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;instruction&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token punctuation\">(<\/span><br \/>\n                <span class=\"token string\">&#034;\u6839\u636e\u5f53\u524d\u753b\u9762\u9009\u62e9\u4e0b\u4e00\u6b65\u52a8\u4f5c\u3002&#034;<\/span><br \/>\n                <span class=\"token string\">&#034;\u8f93\u51fa\u4e00\u4e2a JSON \u5bf9\u8c61&#xff0c;\u5305\u542b action \u548c\u7b80\u77ed\u7684\u53ef\u9a8c\u8bc1\u4f9d\u636e\u3002&#034;<\/span><br \/>\n            <span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        ensure_ascii<span class=\"token operator\">&#061;<\/span><span class=\"token boolean\">False<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">)<\/span><\/p>\n<p>response <span class=\"token operator\">&#061;<\/span> client<span class=\"token punctuation\">.<\/span>responses<span class=\"token punctuation\">.<\/span>create<span class=\"token punctuation\">(<\/span><br \/>\n    model<span class=\"token operator\">&#061;<\/span>model<span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token builtin\">input<\/span><span class=\"token operator\">&#061;<\/span>build_observation<span class=\"token punctuation\">(<\/span><br \/>\n        <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#034;player&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">1<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#034;objects&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token punctuation\">[<\/span><span class=\"token number\">2<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">1<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#034;blue_block&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n<span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token keyword\">for<\/span> step <span class=\"token keyword\">in<\/span> <span class=\"token builtin\">range<\/span><span class=\"token punctuation\">(<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">20<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token comment\"># \u5b9e\u9645\u9879\u76ee\u4e2d\u5e94\u6821\u9a8c\u6a21\u578b\u52a8\u4f5c&#xff0c;\u518d\u4ea4\u7ed9\u73af\u5883\u6267\u884c\u3002<\/span><br \/>\n    action_text <span class=\"token operator\">&#061;<\/span> response<span class=\"token punctuation\">.<\/span>output_text<br \/>\n    next_frame <span class=\"token operator\">&#061;<\/span> environment_step<span class=\"token punctuation\">(<\/span>action_text<span class=\"token punctuation\">)<\/span>  <span class=\"token comment\"># \u7531\u4f60\u7684\u73af\u5883\u5b9e\u73b0<\/span><\/p>\n<p>    response <span class=\"token operator\">&#061;<\/span> client<span class=\"token punctuation\">.<\/span>responses<span class=\"token punctuation\">.<\/span>create<span class=\"token punctuation\">(<\/span><br \/>\n        model<span class=\"token operator\">&#061;<\/span>model<span class=\"token punctuation\">,<\/span><br \/>\n        previous_response_id<span class=\"token operator\">&#061;<\/span>response<span class=\"token punctuation\">.<\/span><span class=\"token builtin\">id<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token builtin\">input<\/span><span class=\"token operator\">&#061;<\/span>build_observation<span class=\"token punctuation\">(<\/span>step<span class=\"token punctuation\">,<\/span> next_frame<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span>response<span class=\"token punctuation\">.<\/span>output_text<span class=\"token punctuation\">)<\/span><\/p>\n<p>\u8fd9\u91cc\u7684\u5173\u952e\u4e0d\u662f\u628a\u5b8c\u6574\u5386\u53f2\u624b\u52a8\u62fc\u56de input&#xff0c;\u800c\u662f\u8ba9\u540e\u7eed\u8bf7\u6c42\u5f15\u7528\u4e0a\u4e00\u8f6e\u54cd\u5e94\u3002\u5b98\u65b9\u5b9e\u9a8c\u79f0&#xff0c;\u5bf9\u4e8e GPT-5.6&#xff0c;\u8fd9\u79cd\u65b9\u5f0f\u53ef\u4ee5\u5728\u5de5\u5177\u8c03\u7528\u548c\u591a\u8f6e\u5bf9\u8bdd\u4e4b\u95f4\u81ea\u52a8\u4fdd\u7559\u63a8\u7406\u72b6\u6001\u3002<\/p>\n<p>\u793a\u4f8b\u4e2d\u7684 environment_step() \u662f\u5360\u4f4d\u51fd\u6570\u3002\u751f\u4ea7\u73af\u5883\u5fc5\u987b\u5728\u5b83\u524d\u9762\u589e\u52a0\u52a8\u4f5c Schema \u6821\u9a8c\u3001\u6743\u9650\u63a7\u5236\u3001\u8d85\u65f6\u3001\u5e42\u7b49\u952e\u548c\u5931\u8d25\u6062\u590d&#xff0c;\u4e0d\u80fd\u76f4\u63a5\u628a\u6a21\u578b\u8f93\u51fa\u5f53\u4f5c\u53ef\u4fe1\u547d\u4ee4\u6267\u884c\u3002<\/p>\n<h3>Compaction \u4e0e\u7c97\u66b4\u622a\u65ad\u6709\u4ec0\u4e48\u4e0d\u540c<\/h3>\n<p>\u6eda\u52a8\u622a\u65ad\u7684\u903b\u8f91\u5f88\u7b80\u5355&#xff1a;\u4e0a\u4e0b\u6587\u8d85\u8fc7\u9650\u5236\u540e&#xff0c;\u4ece\u6700\u65e9\u7684\u6d88\u606f\u5f00\u59cb\u5220\u9664\u3002<\/p>\n<p><span class=\"token keyword\">while<\/span> estimate_tokens<span class=\"token punctuation\">(<\/span>history<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&gt;<\/span> context_limit<span class=\"token punctuation\">:<\/span><br \/>\n    history<span class=\"token punctuation\">.<\/span>pop<span class=\"token punctuation\">(<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<p>\u8fd9\u79cd\u65b9\u5f0f\u5bb9\u6613\u5b9e\u73b0&#xff0c;\u5374\u4e0d\u77e5\u9053\u88ab\u5220\u9664\u7684\u4fe1\u606f\u662f\u5426\u91cd\u8981\u3002\u6700\u65e9\u7684\u5185\u5bb9\u53ef\u80fd\u53ea\u662f\u5bd2\u6684&#xff0c;\u4e5f\u53ef\u80fd\u662f\u4efb\u52a1\u76ee\u6807\u3001\u5df2\u7ecf\u9a8c\u8bc1\u7684\u89c4\u5219\u6216\u4e00\u6b21\u5173\u952e\u5931\u8d25\u3002<\/p>\n<p>Compaction \u7684\u76ee\u6807\u662f\u628a\u8f83\u957f\u5386\u53f2\u538b\u7f29\u6210\u66f4\u7d27\u51d1\u7684\u72b6\u6001&#xff0c;\u518d\u8ba9\u4efb\u52a1\u7ee7\u7eed\u8fd0\u884c\u3002\u6982\u5ff5\u4e0a\u66f4\u63a5\u8fd1\u4e0b\u9762\u7684\u8fc7\u7a0b&#xff1a;<\/p>\n<p><span class=\"token keyword\">if<\/span> estimate_tokens<span class=\"token punctuation\">(<\/span>history<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&gt;<\/span> compact_threshold<span class=\"token punctuation\">:<\/span><br \/>\n    history <span class=\"token operator\">&#061;<\/span> compact<span class=\"token punctuation\">(<\/span><br \/>\n        history<span class=\"token punctuation\">,<\/span><br \/>\n        preserve<span class=\"token operator\">&#061;<\/span><span class=\"token punctuation\">[<\/span><br \/>\n            <span class=\"token string\">&#034;\u4efb\u52a1\u76ee\u6807&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;\u5df2\u9a8c\u8bc1\u89c4\u5219&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;\u5931\u8d25\u5c1d\u8bd5&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;\u672a\u89e3\u51b3\u95ee\u9898&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;\u5916\u90e8\u5de5\u5177\u8fd4\u56de\u7684\u5173\u952e\u4e8b\u5b9e&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">)<\/span><\/p>\n<p>\u8fd9\u6bb5\u4ee3\u7801\u662f\u673a\u5236\u793a\u610f&#xff0c;\u4e0d\u662f OpenAI SDK \u7684\u53c2\u6570\u683c\u5f0f\u3002\u5177\u4f53 Compaction \u914d\u7f6e\u5e94\u4ee5\u6587\u7ae0\u5f00\u5934\u94fe\u63a5\u7684\u5f53\u524d\u5b98\u65b9\u6307\u5357\u4e3a\u51c6\u3002<\/p>\n<p>\u538b\u7f29\u4e5f\u4e0d\u662f\u65e0\u635f\u5b58\u50a8\u3002\u6458\u8981\u53ef\u80fd\u9057\u6f0f\u7cbe\u786e\u6570\u503c\u3001\u4ee3\u7801\u7247\u6bb5\u3001\u5bf9\u8c61 ID \u6216\u8fb9\u754c\u6761\u4ef6\u3002\u56e0\u6b64&#xff0c;\u751f\u4ea7\u7cfb\u7edf\u4e0d\u5e94\u53ea\u4fdd\u7559\u538b\u7f29\u540e\u7684\u81ea\u7136\u8bed\u8a00&#xff0c;\u8fd8\u5e94\u628a\u4e0d\u53ef\u4e22\u5931\u7684\u72b6\u6001\u653e\u5728\u7ed3\u6784\u5316\u5b58\u50a8\u4e2d&#xff1a;<\/p>\n<p><span class=\"token punctuation\">{<\/span><br \/>\n  <span class=\"token string-property property\">&#034;goal&#034;<\/span><span class=\"token operator\">:<\/span> <span class=\"token string\">&#034;\u901a\u8fc7\u7b2c 4 \u5173&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n  <span class=\"token string-property property\">&#034;verified_rules&#034;<\/span><span class=\"token operator\">:<\/span> <span class=\"token punctuation\">[<\/span><br \/>\n    <span class=\"token string\">&#034;blue_block_pushable&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token string\">&#034;red_tile_resets_level&#034;<\/span><br \/>\n  <span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span><br \/>\n  <span class=\"token string-property property\">&#034;completed_steps&#034;<\/span><span class=\"token operator\">:<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">2<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">3<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span><br \/>\n  <span class=\"token string-property property\">&#034;open_questions&#034;<\/span><span class=\"token operator\">:<\/span> <span class=\"token punctuation\">[<\/span><br \/>\n    <span class=\"token string\">&#034;whether_both_switches_must_remain_active&#034;<\/span><br \/>\n  <span class=\"token punctuation\">]<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p>\u6a21\u578b\u4e0a\u4e0b\u6587\u9002\u5408\u4fdd\u5b58\u5de5\u4f5c\u8bb0\u5fc6&#xff0c;\u6570\u636e\u5e93\u6216\u72b6\u6001\u673a\u9002\u5408\u4fdd\u5b58\u4e0d\u80fd\u4e22\u5931\u7684\u4e8b\u5b9e\u3002\u4e24\u8005\u4e0d\u662f\u4e92\u76f8\u66ff\u4ee3\u7684\u5173\u7cfb\u3002<\/p>\n<h3>\u5982\u4f55\u7ed9\u81ea\u5df1\u7684 Agent \u505a Harness A\/B \u6d4b\u8bd5<\/h3>\n<p>\u4ec5\u6bd4\u8f83\u6700\u7ec8\u6210\u529f\u7387\u901a\u5e38\u4e0d\u591f\u3002\u4e00\u4e2a Harness \u53ef\u80fd\u5206\u6570\u66f4\u9ad8&#xff0c;\u4f46\u6d88\u8017\u66f4\u591a Token\u3001\u5ef6\u8fdf\u66f4\u957f&#xff0c;\u6216\u8005\u4f9d\u8d56\u66f4\u591a\u91cd\u8bd5\u3002<\/p>\n<p>\u5efa\u8bae\u81f3\u5c11\u8bb0\u5f55\u4ee5\u4e0b\u5b57\u6bb5&#xff1a;<\/p>\n<p>run_id<br \/>\ntask_id<br \/>\nharness<br \/>\nsuccess<br \/>\nscore<br \/>\ninput_tokens<br \/>\noutput_tokens<br \/>\nlatency_seconds<br \/>\ntool_calls<br \/>\nretries<br \/>\ncompaction_events<\/p>\n<p>\u53ef\u4ee5\u628a\u6bcf\u6b21\u8fd0\u884c\u4fdd\u5b58\u4e3a results.csv&#xff0c;\u518d\u7528\u4e0b\u9762\u7684 Python \u811a\u672c\u505a\u57fa\u7840\u6c47\u603b&#xff1a;<\/p>\n<p><span class=\"token keyword\">from<\/span> collections <span class=\"token keyword\">import<\/span> defaultdict<br \/>\n<span class=\"token keyword\">import<\/span> csv<\/p>\n<p>groups <span class=\"token operator\">&#061;<\/span> defaultdict<span class=\"token punctuation\">(<\/span><span class=\"token builtin\">list<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token keyword\">with<\/span> <span class=\"token builtin\">open<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;results.csv&#034;<\/span><span class=\"token punctuation\">,<\/span> encoding<span class=\"token operator\">&#061;<\/span><span class=\"token string\">&#034;utf-8&#034;<\/span><span class=\"token punctuation\">,<\/span> newline<span class=\"token operator\">&#061;<\/span><span class=\"token string\">&#034;&#034;<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token keyword\">as<\/span> <span class=\"token builtin\">file<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    <span class=\"token keyword\">for<\/span> row <span class=\"token keyword\">in<\/span> csv<span class=\"token punctuation\">.<\/span>DictReader<span class=\"token punctuation\">(<\/span><span class=\"token builtin\">file<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n        groups<span class=\"token punctuation\">[<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;harness&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>append<span class=\"token punctuation\">(<\/span><br \/>\n            <span class=\"token punctuation\">{<\/span><br \/>\n                <span class=\"token string\">&#034;success&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;success&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                <span class=\"token string\">&#034;score&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">float<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;score&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                <span class=\"token string\">&#034;input_tokens&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;input_tokens&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                <span class=\"token string\">&#034;output_tokens&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;output_tokens&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                <span class=\"token string\">&#034;latency&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">float<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;latency_seconds&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                <span class=\"token string\">&#034;retries&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">int<\/span><span class=\"token punctuation\">(<\/span>row<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;retries&#034;<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token punctuation\">}<\/span><br \/>\n        <span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token keyword\">for<\/span> harness<span class=\"token punctuation\">,<\/span> runs <span class=\"token keyword\">in<\/span> <span class=\"token builtin\">sorted<\/span><span class=\"token punctuation\">(<\/span>groups<span class=\"token punctuation\">.<\/span>items<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span><br \/>\n    count <span class=\"token operator\">&#061;<\/span> <span class=\"token builtin\">len<\/span><span class=\"token punctuation\">(<\/span>runs<span class=\"token punctuation\">)<\/span><br \/>\n    total_tokens <span class=\"token operator\">&#061;<\/span> <span class=\"token builtin\">sum<\/span><span class=\"token punctuation\">(<\/span><br \/>\n        run<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;input_tokens&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token operator\">&#043;<\/span> run<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;output_tokens&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token keyword\">for<\/span> run <span class=\"token keyword\">in<\/span> runs<br \/>\n    <span class=\"token punctuation\">)<\/span><\/p>\n<p>    <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span><br \/>\n        <span class=\"token punctuation\">{<\/span><br \/>\n            <span class=\"token string\">&#034;harness&#034;<\/span><span class=\"token punctuation\">:<\/span> harness<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;runs&#034;<\/span><span class=\"token punctuation\">:<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;success_rate&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">sum<\/span><span class=\"token punctuation\">(<\/span>r<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;success&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token keyword\">for<\/span> r <span class=\"token keyword\">in<\/span> runs<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;avg_score&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">sum<\/span><span class=\"token punctuation\">(<\/span>r<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;score&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token keyword\">for<\/span> r <span class=\"token keyword\">in<\/span> runs<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;avg_tokens&#034;<\/span><span class=\"token punctuation\">:<\/span> total_tokens <span class=\"token operator\">\/<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;avg_latency_s&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">sum<\/span><span class=\"token punctuation\">(<\/span>r<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;latency&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token keyword\">for<\/span> r <span class=\"token keyword\">in<\/span> runs<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n            <span class=\"token string\">&#034;avg_retries&#034;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token builtin\">sum<\/span><span class=\"token punctuation\">(<\/span>r<span class=\"token punctuation\">[<\/span><span class=\"token string\">&#034;retries&#034;<\/span><span class=\"token punctuation\">]<\/span> <span class=\"token keyword\">for<\/span> r <span class=\"token keyword\">in<\/span> runs<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">\/<\/span> count<span class=\"token punctuation\">,<\/span><br \/>\n        <span class=\"token punctuation\">}<\/span><br \/>\n    <span class=\"token punctuation\">)<\/span><\/p>\n<p>\u6d4b\u8bd5\u65f6\u5e94\u4fdd\u8bc1\u4e24\u7ec4\u4f7f\u7528\u76f8\u540c\u7684\u4efb\u52a1\u3001\u6a21\u578b\u7248\u672c\u3001\u63a8\u7406\u5f3a\u5ea6\u3001\u5de5\u5177\u6743\u9650\u548c\u6700\u5927\u8fd0\u884c\u6b65\u6570\u3002\u82e5\u4e00\u7ec4\u542f\u7528\u4e86\u72b6\u6001\u5ef6\u7eed&#xff0c;\u53e6\u4e00\u7ec4\u6ca1\u6709&#xff0c;\u90a3\u4e48\u6d4b\u8bd5\u95ee\u9898\u5e94\u660e\u786e\u5199\u6210\u201c\u54ea\u5957 Agent \u7cfb\u7edf\u5b8c\u6210\u4efb\u52a1\u66f4\u597d\u201d&#xff0c;\u800c\u4e0d\u662f\u201c\u54ea\u4e2a\u6a21\u578b\u66f4\u5f3a\u201d\u3002<\/p>\n<h3>\u516c\u5e73\u8bc4\u6d4b\u4e0e\u771f\u5b9e\u4ea7\u54c1\u8bc4\u6d4b\u5e76\u4e0d\u5b8c\u5168\u76f8\u540c<\/h3>\n<p>ARC-AGI-3 \u91c7\u7528\u901a\u7528 Harness \u6709\u5408\u7406\u76ee\u7684&#xff1a;\u51cf\u5c11\u9488\u5bf9\u7279\u5b9a\u6a21\u578b\u7684\u9002\u914d&#xff0c;\u8ba9\u4e0d\u540c\u6a21\u578b\u5728\u66f4\u4e00\u81f4\u7684\u6761\u4ef6\u4e0b\u6bd4\u8f83\u3002OpenAI \u4f7f\u7528\u751f\u4ea7\u5316 Harness \u4e5f\u6709\u5408\u7406\u76ee\u7684&#xff1a;\u6d4b\u91cf\u6a21\u578b\u5728\u5b9e\u9645 API \u80fd\u529b\u88ab\u6b63\u786e\u4f7f\u7528\u65f6\u53ef\u4ee5\u505a\u5230\u4ec0\u4e48\u3002<\/p>\n<p>\u4e24\u79cd\u8bc4\u6d4b\u56de\u7b54\u7684\u662f\u4e0d\u540c\u95ee\u9898&#xff1a;<\/p>\n<table>\n<tr>\u8bc4\u6d4b\u65b9\u5f0f\u66f4\u63a5\u8fd1\u7684\u95ee\u9898<\/tr>\n<tbody>\n<tr>\n<td>\u901a\u7528 Harness<\/td>\n<td>\u5728\u76f8\u540c\u3001\u7b80\u5316\u7684\u5916\u90e8\u6761\u4ef6\u4e0b&#xff0c;\u6a21\u578b\u672c\u8eab\u8868\u73b0\u5982\u4f55<\/td>\n<\/tr>\n<tr>\n<td>\u6a21\u578b\u9002\u914d Harness<\/td>\n<td>\u4f7f\u7528\u5382\u5546\u63a8\u8350\u914d\u7f6e\u540e&#xff0c;\u5b8c\u6574\u7cfb\u7edf\u80fd\u8fbe\u5230\u4ec0\u4e48\u7ed3\u679c<\/td>\n<\/tr>\n<tr>\n<td>\u4e1a\u52a1 Harness<\/td>\n<td>\u5728\u771f\u5b9e\u6743\u9650\u3001\u6210\u672c\u3001\u5ef6\u8fdf\u548c\u5931\u8d25\u7ea6\u675f\u4e0b&#xff0c;\u4efb\u52a1\u80fd\u5426\u7a33\u5b9a\u5b8c\u6210<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u56e0\u6b64&#xff0c;\u770b\u5230\u201c\u5f00\u542f\u4e24\u4e2a\u8bbe\u7f6e\u540e\u5f97\u5206\u63a5\u8fd1\u4e09\u500d\u201d\u65f6&#xff0c;\u4e0d\u80fd\u76f4\u63a5\u5f97\u51fa\u539f\u699c\u5355\u65e0\u6548&#xff0c;\u4e5f\u4e0d\u80fd\u628a 38.3% \u5f53\u4f5c\u6240\u6709 Agent \u4efb\u52a1\u90fd\u4f1a\u83b7\u5f97\u7684\u56fa\u5b9a\u63d0\u5347\u3002<\/p>\n<p>\u66f4\u7a33\u59a5\u7684\u505a\u6cd5\u662f\u540c\u65f6\u62a5\u544a&#xff1a;\u6a21\u578b\u540d\u79f0\u3001API \u7248\u672c\u3001\u63a8\u7406\u8bbe\u7f6e\u3001\u4e0a\u4e0b\u6587\u7b56\u7565\u3001\u5de5\u5177\u96c6\u5408\u3001\u505c\u6b62\u6761\u4ef6\u3001Token \u4f7f\u7528\u91cf\u548c\u5931\u8d25\u6837\u672c\u3002\u7f3a\u5c11\u8fd9\u4e9b\u4fe1\u606f&#xff0c;\u5355\u72ec\u4e00\u4e2a\u5206\u6570\u5f88\u96be\u590d\u73b0\u3002<\/p>\n<h3>\u751f\u4ea7 Agent \u8fd8\u9700\u8981\u54ea\u4e9b\u4fdd\u62a4<\/h3>\n<p>\u4fdd\u7559\u63a8\u7406\u548c\u538b\u7f29\u4e3b\u8981\u89e3\u51b3\u8bb0\u5fc6\u8fde\u7eed\u6027&#xff0c;\u4e0d\u4f1a\u81ea\u52a8\u89e3\u51b3\u5de5\u5177\u5b89\u5168\u548c\u7ed3\u679c\u6b63\u786e\u6027\u3002\u957f\u7a0b Agent \u81f3\u5c11\u8fd8\u9700\u8981&#xff1a;<\/p>\n<ul>\n<li>\u7ed9\u5de5\u5177\u8c03\u7528\u5b9a\u4e49\u4e25\u683c Schema&#xff0c;\u62d2\u7edd\u672a\u77e5\u5b57\u6bb5\u548c\u8d8a\u754c\u53c2\u6570&#xff1b;<\/li>\n<li>\u5bf9\u5199\u6587\u4ef6\u3001\u53d1\u6d88\u606f\u3001\u4ed8\u6b3e\u548c\u5220\u9664\u64cd\u4f5c\u589e\u52a0\u4eba\u5de5\u786e\u8ba4&#xff1b;<\/li>\n<li>\u7ed9\u5916\u90e8\u52a8\u4f5c\u8bbe\u7f6e\u5e42\u7b49\u952e&#xff0c;\u907f\u514d\u7f51\u7edc\u91cd\u8bd5\u9020\u6210\u91cd\u590d\u6267\u884c&#xff1b;<\/li>\n<li>\u5206\u522b\u8bb0\u5f55\u6a21\u578b\u5931\u8d25\u3001\u5de5\u5177\u5931\u8d25\u548c\u4e1a\u52a1\u9a8c\u6536\u5931\u8d25&#xff1b;<\/li>\n<li>\u4fdd\u5b58\u73af\u5883\u4e2d\u7684\u6743\u5a01\u72b6\u6001&#xff0c;\u800c\u4e0d\u662f\u53ea\u4f9d\u8d56\u5bf9\u8bdd\u5386\u53f2&#xff1b;<\/li>\n<li>\u5bf9\u538b\u7f29\u524d\u540e\u7684\u5173\u952e\u4e8b\u5b9e\u505a\u4e00\u81f4\u6027\u68c0\u67e5&#xff1b;<\/li>\n<li>\u8bbe\u7f6e Token\u3001\u65f6\u95f4\u3001\u6b65\u9aa4\u6570\u548c\u8d39\u7528\u9884\u7b97&#xff1b;<\/li>\n<li>\u4fdd\u7559\u5b8c\u6574\u8fd0\u884c\u8f68\u8ff9&#xff0c;\u4fbf\u4e8e\u5b9a\u4f4d\u67d0\u4e00\u6b65\u4e3a\u4ec0\u4e48\u504f\u79bb\u76ee\u6807\u3002<\/li>\n<\/ul>\n<p>\u4e0a\u4e0b\u6587\u7ba1\u7406\u505a\u5f97\u66f4\u597d&#xff0c;Agent \u901a\u5e38\u80fd\u8fd0\u884c\u5f97\u66f4\u4e45&#xff1b;\u8fd0\u884c\u65f6\u95f4\u66f4\u957f&#xff0c;\u4e5f\u610f\u5473\u7740\u9519\u8bef\u53ef\u80fd\u4f20\u64ad\u5f97\u66f4\u8fdc\u3002\u56e0\u6b64&#xff0c;\u8bb0\u5fc6\u80fd\u529b\u4e0e\u6743\u9650\u63a7\u5236\u9700\u8981\u4e00\u8d77\u8bbe\u8ba1\u3002<\/p>\n<h3>\u8fd9\u9879\u5b9e\u9a8c\u7ed9\u5f00\u53d1\u8005\u7684\u5b9e\u9645\u542f\u53d1<\/h3>\n<p>ARC-AGI-3 \u5b9e\u9a8c\u8bf4\u660e&#xff0c;\u6a21\u578b\u80fd\u529b\u53ea\u662f Agent \u7cfb\u7edf\u7684\u4e00\u90e8\u5206\u3002\u5bf9\u4e8e\u9700\u8981\u8fde\u7eed\u5de5\u5177\u8c03\u7528\u3001\u4ee3\u7801\u4fee\u6539\u3001\u6e38\u620f\u64cd\u4f5c\u3001\u7f51\u9875\u7814\u7a76\u6216\u79d1\u5b66\u8ba1\u7b97\u7684\u957f\u4efb\u52a1&#xff0c;Harness \u81f3\u5c11\u627f\u62c5\u56db\u9879\u804c\u8d23&#xff1a;<\/p>\n<li>\u4fdd\u7559\u4efb\u52a1\u76ee\u6807\u548c\u5df2\u7ecf\u5f62\u6210\u7684\u8ba1\u5212&#xff1b;<\/li>\n<li>\u63a7\u5236\u4e0a\u4e0b\u6587\u589e\u957f&#xff0c;\u5e76\u907f\u514d\u4e22\u5931\u5173\u952e\u4e8b\u5b9e&#xff1b;<\/li>\n<li>\u5c06\u6a21\u578b\u52a8\u4f5c\u5b89\u5168\u5730\u8f6c\u6362\u6210\u73af\u5883\u64cd\u4f5c&#xff1b;<\/li>\n<li>\u7528\u4e1a\u52a1\u6307\u6807\u8861\u91cf\u7ed3\u679c&#xff0c;\u800c\u4e0d\u53ea\u7edf\u8ba1\u56de\u7b54\u8d28\u91cf\u3002<\/li>\n<p>OpenAI \u7684\u7ed3\u679c\u6765\u81ea\u5b98\u65b9\u5b9e\u9a8c&#xff0c;\u5c1a\u4e0d\u80fd\u66ff\u4ee3\u4e0d\u540c\u6a21\u578b\u3001\u4e0d\u540c\u5de5\u5177\u548c\u771f\u5b9e\u4e1a\u52a1\u4e0a\u7684\u72ec\u7acb\u590d\u73b0\u3002\u4f46\u5b83\u63d0\u4f9b\u4e86\u4e00\u4e2a\u503c\u5f97\u91c7\u7528\u7684\u8bc4\u6d4b\u4e60\u60ef&#xff1a;\u6a21\u578b\u5347\u7ea7\u524d&#xff0c;\u5148\u628a Harness \u914d\u7f6e\u3001\u4e0a\u4e0b\u6587\u7b56\u7565\u548c\u5de5\u5177\u6743\u9650\u5199\u8fdb\u5b9e\u9a8c\u8bb0\u5f55&#xff1b;\u770b\u5230\u5f02\u5e38\u4f4e\u5206\u65f6&#xff0c;\u4e5f\u5148\u68c0\u67e5\u72b6\u6001\u662f\u5426\u5728\u8f6e\u6b21\u95f4\u88ab\u610f\u5916\u4e22\u5f03\u3002<\/p>\n<p>Agent \u65f6\u4ee3\u7684\u6a21\u578b\u8bc4\u6d4b&#xff0c;\u6b63\u5728\u4ece\u201c\u7ed9\u6a21\u578b\u4e00\u5f20\u8bd5\u5377\u201d\u53d8\u6210\u201c\u7ed9\u7cfb\u7edf\u4e00\u9879\u8fde\u7eed\u4efb\u52a1\u201d\u3002\u8fd9\u4e24\u79cd\u6d4b\u8bd5\u90fd\u91cd\u8981&#xff0c;\u53ea\u662f\u4e0d\u80fd\u6df7\u4e3a\u4e00\u8c08\u3002<\/p>\n<h3>\u53c2\u8003\u8d44\u6599<\/h3>\n<ul>\n<li>OpenAI&#xff1a;How enabling two settings tripled our scores on the ARC-AGI-3 benchmark&#xff0c;2026-07-29<br \/>\nhttps:\/\/openai.com\/index\/how-two-settings-tripled-our-arc-agi-3-scores\/<\/li>\n<li>OpenAI Developers&#xff1a;Responses API<br \/>\nhttps:\/\/developers.openai.com\/blog\/responses-api<\/li>\n<li>OpenAI Developers&#xff1a;Compaction<br \/>\nhttps:\/\/developers.openai.com\/api\/docs\/guides\/compaction<\/li>\n<li>ARC Prize&#xff1a;ARC-AGI-3<br \/>\nhttps:\/\/arcprize.org\/arc-agi\/3<\/li>\n<li>ARC Prize Docs&#xff1a;RHAE Methodology<br \/>\nhttps:\/\/docs.arcprize.org\/methodology<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u8fd1\u671fAI\u70ed\u70b9007&#xff5c;ARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d&#xff1a;\u6a21\u578b\u4e4b\u5916&#xff0c;Agent Harness \u4e5f\u662f\u53d8\u91cf\u4e3b\u8981\u4fe1\u606f\u6e90&#xff1a;OpenAI ARC-AGI-3 \u5b98\u65b9\u5b9e\u9a8c\u6587\u7ae0\u3001Responses API \u5b98\u65b9\u4ecb\u7ecd\u3001Compaction \u5b98\u65b9\u6307\u5357\u3001ARC Prize \u65b9\u6cd5\u8bf4\u660e<br \/>\n\u5173\u952e\u8bcd&#xff1a;ARC-AGI-3\u3001Agent Harness\u3001Responses API\u3001Compaction\u3001\u4e0a\u4e0b\u6587\u7ba1\u7406\u3001AI Agent\u3001\u6a21\u578b\u8bc4<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[10545,10546,1699,71,50],"topic":[],"class_list":["post-92234","post","type-post","status-publish","format-standard","hentry","category-server","tag-agent-harness","tag-responses-api","tag-agi","tag-ai-agent","tag-50"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \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\/92234.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u8fd1\u671fAI\u70ed\u70b9007&#xff5c;ARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d&#xff1a;\u6a21\u578b\u4e4b\u5916&#xff0c;Agent Harness \u4e5f\u662f\u53d8\u91cf\u4e3b\u8981\u4fe1\u606f\u6e90&#xff1a;OpenAI ARC-AGI-3 \u5b98\u65b9\u5b9e\u9a8c\u6587\u7ae0\u3001Responses API \u5b98\u65b9\u4ecb\u7ecd\u3001Compaction \u5b98\u65b9\u6307\u5357\u3001ARC Prize \u65b9\u6cd5\u8bf4\u660e \u5173\u952e\u8bcd&#xff1a;ARC-AGI-3\u3001Agent Harness\u3001Responses API\u3001Compaction\u3001\u4e0a\u4e0b\u6587\u7ba1\u7406\u3001AI Agent\u3001\u6a21\u578b\u8bc4\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/92234.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-09T11:56:41+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/92234.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/92234.html\",\"name\":\"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-09T11:56:41+00:00\",\"dateModified\":\"2026-08-09T11:56:41+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/92234.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/92234.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/92234.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf\"}]},{\"@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":"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \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\/92234.html","og_locale":"zh_CN","og_type":"article","og_title":"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u8fd1\u671fAI\u70ed\u70b9007&#xff5c;ARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d&#xff1a;\u6a21\u578b\u4e4b\u5916&#xff0c;Agent Harness \u4e5f\u662f\u53d8\u91cf\u4e3b\u8981\u4fe1\u606f\u6e90&#xff1a;OpenAI ARC-AGI-3 \u5b98\u65b9\u5b9e\u9a8c\u6587\u7ae0\u3001Responses API \u5b98\u65b9\u4ecb\u7ecd\u3001Compaction \u5b98\u65b9\u6307\u5357\u3001ARC Prize \u65b9\u6cd5\u8bf4\u660e \u5173\u952e\u8bcd&#xff1a;ARC-AGI-3\u3001Agent Harness\u3001Responses API\u3001Compaction\u3001\u4e0a\u4e0b\u6587\u7ba1\u7406\u3001AI Agent\u3001\u6a21\u578b\u8bc4","og_url":"https:\/\/www.wsisp.com\/helps\/92234.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-09T11:56:41+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/92234.html","url":"https:\/\/www.wsisp.com\/helps\/92234.html","name":"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-09T11:56:41+00:00","dateModified":"2026-08-09T11:56:41+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/92234.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/92234.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/92234.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u8fd1\u671fAI\u70ed\u70b9007\uff5cARC-AGI-3 \u5f97\u5206\u4e3a\u4f55\u63d0\u9ad8\u5230\u8fd1\u4e09\u500d\uff1a\u6a21\u578b\u4e4b\u5916\uff0cAgent Harness \u4e5f\u662f\u53d8\u91cf"}]},{"@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\/92234","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=92234"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/92234\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=92234"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=92234"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=92234"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=92234"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}