{"id":764739,"date":"2026-04-19T14:10:22","date_gmt":"2026-04-19T06:10:22","guid":{"rendered":"https:\/\/www.alphaseek.cn\/?p=764739"},"modified":"2026-04-19T14:10:22","modified_gmt":"2026-04-19T06:10:22","slug":"swe-bench%e6%bb%a1%e5%88%86%ef%bc%8c0%e4%b8%aabug%e4%bf%ae%e5%a4%8d%ef%bc%9a%e4%bc%af%e5%85%8b%e5%88%a9%e9%80%a0%e4%ba%86%e4%b8%aa%e4%b8%93%e9%97%a8%e4%bd%9c%e5%bc%8a%e7%9a%84ai","status":"publish","type":"post","link":"https:\/\/www.alphaseek.cn\/?p=764739","title":{"rendered":"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI"},"content":{"rendered":"<p><strong>\u65b0\u667a\u5143\u62a5\u9053<\/strong><\/p>\n<p>\u7f16\u8f91\uff1a\u503e\u503e<\/p>\n<p><strong>\u3010\u65b0\u667a\u5143\u5bfc\u8bfb\u3011\u4f2f\u514b\u5229\u56e2\u961f\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\uff0c\u752810\u884cPython\u4ee3\u7801\u62ff\u4e0bSWE-bench\u6ee1\u5206\uff01500\u9053\u9898\u5168\u8fc7\uff0c0\u4e2abug\u4fee\u590d\u30028\u5927\u4e3b\u6d41\u8bc4\u6d4b\u57fa\u51c6\uff0c\u5168\u90e8\u6ca6\u9677\u3002\u540c\u4e00\u5468\uff0c\u4e24\u4efd\u72ec\u7acb\u5ba1\u8ba1\u786e\u8ba4\uff1a\u6392\u884c\u699c\u4e0a\u7684\u4f5c\u5f0a\u65e9\u5df2\u4e0d\u662f\u5047\u8bbe\uff0c\u800c\u662f\u73b0\u5b9e\u3002<\/strong><\/p>\n<p>\u672c\u5468\uff0cAI\u8bc4\u6d4b\u5708\u7ecf\u5386\u4e86\u4e00\u573a\u4fe1\u4efb\u5730\u9707\u3002<\/p>\n<p>SWE-bench\uff0c\u662f\u516c\u8ba4\u7684AI\u7f16\u7a0b\u80fd\u529b\u6807\u6746\uff0c\u5404\u5927\u6a21\u578b\u53d1\u5e03\u4f1a\u4e0a\u7684\u5fc5\u62a5\u6570\u5b57\uff0c<a href=\"https:\/\/www.alphaseek.cn\/?tag=%e6%8a%95%e8%b5%84\" title=\"View all posts in \u6295\u8d44\" target=\"_blank\" class=\"tag_link\">\u6295\u8d44<\/a>\u4eba\u4f30\u503c\u65f6\u7684\u786c\u901a\u8d27\u3002<\/p>\n<p>\u53ef\u4f2f\u514b\u5229\u7684\u7814\u7a76\u56e2\u961f\u544a\u8bc9\u4f60\uff0c\u4e00\u4e2aconftest.py\u6587\u4ef6\u5c31\u80fd\u8ba9\u5b83\u7834\u9632\uff01<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/7BDAAC50205F96E32FF913B1ADFEB6C9FCBD1751_size143_w1080_h633.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>\u4e0d\u53eaSWE-bench\u3002\u4f2f\u514b\u5229RDI\u56e2\u961f\u9020\u4e86\u4e00\u4e2a\u81ea\u52a8\u5316\u6f0f\u6d1e\u626b\u63cf\u667a\u80fd\u4f53\uff0c\u5bf9\u5f53\u524d\u6700\u4e3b\u6d41\u76848\u4e2aAI\u667a\u80fd\u4f53\u8bc4\u6d4b\u57fa\u51c6\u9010\u4e00\u6e17\u900f\u3002<\/p>\n<p>\u7ed3\u679c\uff0c\u6bcf\u4e00\u4e2a\u90fd\u88ab\u653b\u7834\uff0c\u5f97\u5206\u4ece73%\u5230100%\u4e0d\u7b49\u3002<\/p>\n<p>\u66f4\u5de7\u7684\u662f\uff0c\u540c\u4e00\u5468\uff0c\u5bbe\u5927\u56e2\u961f\u7684\u72ec\u7acb\u5ba1\u8ba1\u62a5\u544a\u548cAnthropic\u7684Mythos Preview\u7cfb\u7edf\u5361\u540c\u65f6\u51fa\u7089\uff0c\u4e09\u6761\u7ebf\u6307\u5411\u540c\u4e00\u4e2a\u7ed3\u8bba\uff1a\u8fd9\u4e9b\u8bc4\u6d4b\u57fa\u51c6\uff0c\u4ece\u8bbe\u8ba1\u5230\u6267\u884c\uff0c\u5230\u5904\u662f\u6f0f\u6d1e\u3002<\/p>\n<p><strong>10\u884c\u4ee3\u7801\uff0c500\u9898\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d<\/strong><\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/F293526993D3F10E5B76DFF1F18D0FFDC5A8CFB9_size114_w1080_h625.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u7684\u6f0f\u6d1e\u5229\u7528\u667a\u80fd\u4f53\u57288\u5927\u57fa\u51c6\u4e0a\u7684\u5f97\u5206\u3002\u6ca1\u6709\u89e3\u51b3\u4efb\u4f55\u4efb\u52a1\uff0c\u6ca1\u6709\u8c03\u7528\u4efb\u4f55\u5927\u6a21\u578b\uff0c6\u4e2a\u62ff\u5230\u6ee1\u5206\u3002<\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u7684\u7834\u6cd5\uff0c\u7b80\u5355\u5230\u79bb\u8c31\u3002<\/p>\n<p>SWE-bench\u8981\u6c42AI\u4fee\u590d\u771f\u5b9e\u7684GitHub bug\uff0c\u6d4b\u8bd5\u901a\u8fc7\u624d\u7b97\u6210\u529f\u3002<\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u5199\u4e86\u4e00\u4e2aconftest.py\u6587\u4ef6\uff0c\u5229\u7528pytest\u7684\u94a9\u5b50\u673a\u5236\uff0c\u5728\u6d4b\u8bd5\u8fd0\u884c\u65f6\u62e6\u622a\u6bcf\u4e00\u4e2a\u6d4b\u8bd5\u7ed3\u679c\uff0c\u5f3a\u5236\u6539\u5199\u4e3a\u300c\u901a\u8fc7\u300d\u3002<\/p>\n<p>500\u9053\u9898\uff0c\u5168\u90e8\u6ee1\u5206\uff0c\u4e00\u4e2abug\u4e5f\u6ca1\u4fee\u3002<\/p>\n<p>\u5176\u4e2d\u7684\u539f\u7406\u5f88\u7b80\u5355\u3002SWE-bench\u7684\u6d4b\u8bd5\u548c\u88ab\u6d4bAI\u8dd1\u5728<strong>\u540c\u4e00\u4e2aDocker\u5bb9\u5668<\/strong>\u91cc\u3002<\/p>\n<p>\u667a\u80fd\u4f53\u63d0\u4ea4\u7684\u4ee3\u7801\u5728\u5bb9\u5668\u5185\u62e5\u6709\u5b8c\u6574\u6743\u9650\uff0c\u800cpytest\u4f1a\u81ea\u52a8\u53d1\u73b0\u5e76\u52a0\u8f7dconftest.py\u6587\u4ef6\u3002<\/p>\n<p>\u94a9\u5b50\u5728\u6d4b\u8bd5\u7684\u300ccall\u300d\u9636\u6bb5\u62e6\u622a\u7ed3\u679c\uff0c\u628a\u6240\u6709outcome\u6539\u6210passed\u3002<\/p>\n<p>\u65e5\u5fd7\u89e3\u6790\u5668\u770b\u5230\u7684\u662f\u4e00\u7247\u7eff\u706f\uff0c\u8bc4\u5206\u5668\u5224\u5b9a\u5168\u90e8\u901a\u8fc7\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/12A9C5A0A7A9F5604592D890BAEF704ABA85D3E2_size68_w1038_h356.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>SWE-bench\u7684conftest.py\u94a9\u5b50\u6ce8\u5165\u6d41\u7a0b\uff1a\u667a\u80fd\u4f53\u63d0\u4ea4\u7684\u8865\u4e01\u6ca1\u6709\u4fee\u590d\u4efb\u4f55bug\uff0c\u53ea\u585e\u5165\u4e86\u4e00\u4e2aconftest.py\u6587\u4ef6\u3002pytest\u81ea\u52a8\u52a0\u8f7d\u540e\uff0c\u94a9\u5b50\u62e6\u622a\u6bcf\u4e2a\u6d4b\u8bd5\u7ed3\u679c\u5e76\u6539\u5199\u4e3a\u300c\u901a\u8fc7\u300d\u3002<\/p>\n<p>\u5176\u4ed6\u57fa\u51c6\u7684\u7834\u6cd5\u66f4\u52a0\u76f4\u767d\u3002<\/p>\n<p>WebArena\u4efb\u52a1\u7684\u6807\u51c6\u7b54\u6848\u5c31\u5728\u672c\u5730\u7684config_files\u76ee\u5f55\u91cc\uff0cAI\u7528Playwright\u9a71\u52a8\u6d4f\u89c8\u5668\u6253\u5f00\u4e00\u4e2afile:\/\/\u8def\u5f84\uff0c\u76f4\u63a5\u8bfb\u53d6\u7b54\u6848\u3002<\/p>\n<p>\u8bc4\u6d4b\u6846\u67b6\u4ece\u672a\u9650\u5236\u8fc7file:\/\/\u534f\u8bae\u7684\u8bbf\u95ee\u3002\u4e0d\u9700\u8981\u4fee\u6539\u4efb\u4f55\u4ee3\u7801\uff0c\u4e0d\u9700\u8981\u7834\u89e3\u4efb\u4f55\u4e1c\u897f\uff0c\u6d4f\u89c8\u5668\u672c\u8eab\u5c31\u662f\u300c\u7b54\u6848\u9605\u8bfb\u5668\u300d\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/95AFCCA935CF6F2C60939989D845FC7A118460F8_size80_w1002_h310.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>WebArena\u7684file:\/\/\u6f0f\u6d1e\uff1a\u6a21\u578b\u4e0d\u9700\u8981\u6d4f\u89c8\u7f51\u9875\u5b8c\u6210\u4efb\u52a1\uff0c\u53ea\u9700\u8f93\u51fa\u4e00\u6761goto\u6307\u4ee4\uff0cPlaywright\u5c31\u4f1a\u4ece\u672c\u5730\u914d\u7f6e\u6587\u4ef6\u4e2d\u8bfb\u53d6\u6807\u51c6\u7b54\u6848\u5e76\u8fd4\u56de\u3002\u6574\u4e2a\u8fc7\u7a0b\u4e0d\u6d89\u53ca\u4efb\u4f55\u63a8\u7406\u3002<\/p>\n<p>\u6700\u79bb\u8c31\u7684\u662fFieldWorkArena\u3002\u5b83\u7684validate()\u51fd\u6570\u6839\u672c\u4e0d\u68c0\u67e5\u7b54\u6848\u5185\u5bb9\uff0c\u53ea\u770b\u6700\u540e\u4e00\u6761\u6d88\u606f\u662f\u4e0d\u662f\u6765\u81eaassistant\u3002<\/p>\n<p>\u53d1\u4e00\u4e2a\u7a7a\u7684<code>{}<\/code>\uff0c\u5c31\u80fd\u62ff\u6ee1\u5206\u3002<\/p>\n<p>\u90a3\u4e2a\u672c\u5e94\u6bd4\u5bf9\u7b54\u6848\u7684llm_fuzzy_match\u51fd\u6570\uff1f\u5bfc\u5165\u4e86\uff0c\u4f46\u4ece\u672a\u88ab\u8c03\u7528\u3002<\/p>\n<p>\u5269\u4e0b\u7684Terminal-Bench\u3001OSWorld\u3001GAIA\u3001CAR-bench\u3001SWE-bench Pro\uff0c\u624b\u6cd5\u5404\u5f02\u4f46\u903b\u8f91\u76f8\u901a\u3002<\/p>\n<p>\u6728\u9a6c\u5316\u9a8c\u8bc1\u5668\u7684\u4f9d\u8d56\u5de5\u5177\u3001\u4ece\u516c\u5f00URL\u4e0b\u8f7d\u6807\u51c6\u7b54\u6848\u8ba9\u8bc4\u6d4b\u5668\u81ea\u5df1\u548c\u81ea\u5df1\u6bd4\u5bf9\u3001\u5f80LLM\u88c1\u5224\u7684prompt\u91cc\u6ce8\u5165\u9690\u85cf\u6307\u4ee4\u3002<\/p>\n<p>8\u4e2a\u57fa\u51c6\uff0c\u6ca1\u6709\u4e00\u4e2a\u80fd\u62b5\u5fa1\u4e00\u4e2a\u300c\u4ec0\u4e48\u90fd\u4e0d\u4f1a\u4f46\u4e13\u95e8\u627e\u6f0f\u6d1e\u300d\u7684\u667a\u80fd\u4f53\u3002<\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u5f52\u7eb3\u51fa<strong>7\u79cd\u53cd\u590d\u51fa\u73b0\u7684\u6a21\u5f0f<\/strong>\uff1a\u667a\u80fd\u4f53\u548c\u8bc4\u6d4b\u7a0b\u5e8f\u5171\u4eab\u8fd0\u884c\u73af\u5883\u3001\u6807\u51c6\u7b54\u6848\u66b4\u9732\u7ed9\u88ab\u6d4b\u7cfb\u7edf\u3001\u5bf9\u4e0d\u53ef\u4fe1\u8f93\u5165\u8c03\u7528eval()\u3001LLM\u88c1\u5224\u7f3a\u4e4f\u8f93\u5165\u8fc7\u6ee4\u3001\u5b57\u7b26\u4e32\u5339\u914d\u8fc7\u4e8e\u5bbd\u677e\u3001\u8bc4\u5206\u903b\u8f91\u672c\u8eab\u6709bug\u3001\u4ee5\u53ca\u8bc4\u6d4b\u7a0b\u5e8f\u4fe1\u4efb\u88ab\u6d4b\u7cfb\u7edf\u4ea7\u751f\u7684\u8f93\u51fa\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/92D284509D1DC285039F5BA1AB11FB51EB87D8C8_size111_w980_h502.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u5f52\u7eb3\u76847\u79cd\u6f0f\u6d1e\u6a21\u5f0f\u53ca\u5176\u57288\u4e2a\u57fa\u51c6\u4e2d\u7684\u5206\u5e03\u3002\u524d\u4e24\u79cd\uff08\u667a\u80fd\u4f53\u4e0e\u8bc4\u6d4b\u5668\u672a\u9694\u79bb\u3001\u6807\u51c6\u7b54\u6848\u6cc4\u9732\uff09\u51e0\u4e4e\u547d\u4e2d\u4e86\u6240\u6709\u57fa\u51c6\u3002<\/p>\n<p><strong>\u4f5c\u5f0a\uff0c\u6b63\u5728\u53d1\u751f<\/strong><\/p>\n<p>4\u670810\u65e5\uff0c\u5bbe\u5927\u7684Adam Stein\u548cDavis Brown\u53d1\u5e03\u4e86\u4e00\u9879\u5927\u89c4\u6a21\u5ba1\u8ba1\u3002<\/p>\n<p>\u4ed6\u4eec\u7528\u4e00\u4e2a\u53ebMeerkat\u7684\u667a\u80fd\u4f53\u641c\u7d22\u5de5\u5177\uff0c\u626b\u63cf\u4e86\u6570\u5343\u6761\u771f\u5b9e\u7684\u8bc4\u6d4b\u8f68\u8ff9\uff0c\u53d1\u73b0<strong>28+\u4e2a\u63d0\u4ea4\u30019\u4e2a\u57fa\u51c6\u3001\u4e0a\u5343\u6761\u4f5c\u5f0a\u8f68\u8ff9<\/strong>\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/07A6F2073C42ABA0B12D9D259FF9660448CA7915_size124_w1080_h514.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>\u5bbe\u5927Meerkat\u5ba1\u8ba1\u53d1\u73b0\u7684\u4f5c\u5f0a\u6a21\u5f0f\u5206\u5e03\u3002\u6a59\u8272\u4e3aharness\u7ea7\u4f5c\u5f0a\uff08\u5f00\u53d1\u8005\u6846\u67b6\u6cc4\u9732\u7b54\u6848\uff09\uff0c\u84dd\u8272\u4e3a\u4efb\u52a1\u7ea7\u4f5c\u5f0a\uff08\u667a\u80fd\u4f53\u81ea\u884c\u8d70\u6377\u5f84\uff09\u3002\u6ce8\u610f\u6a2a\u8f74\u662f\u5bf9\u6570\u5750\u6807\uff0charness\u7ea7\u4f5c\u5f0a\u7684\u89c4\u6a21\u6bd4\u4efb\u52a1\u7ea7\u9ad8\u51fa\u4e24\u4e2a\u6570\u91cf\u7ea7\u3002<\/p>\n<p>\u6700\u624e\u773c\u7684\u662fTerminal-Bench 2\uff0c\u4e00\u4e2a\u88ab\u7528\u6765\u8bc4\u4f30Opus 4.6\u548cGPT-5.4\u7684\u70ed\u95e8\u57fa\u51c6\u3002\u6392\u884c\u699c\u524d\u4e09\u540d\uff0c\u5168\u90e8\u5b58\u5728\u4f5c\u5f0a\u884c\u4e3a\u3002<\/p>\n<p>\u7b2c\u4e00\u540dPilot\uff0882.9%\u901a\u8fc7\u7387\uff09\uff1a429\u6761\u8f68\u8ff9\u4e2d\uff0c\u6709415\u6761\u7684\u7b2c\u4e00\u4e2a\u52a8\u4f5c\u5c31\u662f<code>cat \/tests\/test_outputs.py<\/code>\uff0c\u8bfb\u53d6\u672c\u5e94\u4e0d\u53ef\u8bbf\u95ee\u7684\u6d4b\u8bd5\u6587\u4ef6\uff0c\u7136\u540e\u53cd\u5411\u63a8\u5bfc\u51fa\u671f\u671b\u8f93\u51fa\u3002<\/p>\n<p>\u7b2c\u4e8c\u540d\u548c\u7b2c\u4e09\u540dForgeCode\uff0881.8%\u901a\u8fc7\u7387\uff09\uff1a\u5b83\u7684harness\u4f1a\u5728\u6267\u884c\u524d\u81ea\u52a8\u52a0\u8f7dAGENTS.md\u6587\u4ef6\u5230\u7cfb\u7edf\u63d0\u793a\u4e2d\uff0c\u800c\u8fd9\u4e9b\u6587\u4ef6\u91cc\u76f4\u63a5\u5305\u542b\u6807\u51c6\u7b54\u6848\u3002<\/p>\n<p>\u4e00\u4e2a\u4efb\u52a1\u91cc\uff0cAGENTS.md\u8d6b\u7136\u5199\u7740\uff1a<\/p>\n<p>\u4e0a\u4e00\u6b21\u8fd0\u884c\u5931\u8d25\u4e86\uff0c\u56e0\u4e3a\u5199\u4e86\u9519\u8bef\u7b54\u6848\u2026\u2026\u6b63\u786e\u7b54\u6848\u5e94\u8be5\u662fGritLM\/GritLM-7B\u3002<\/p>\n<p>\u628aForgeCode\u4e2d\u5f15\u7528AGENTS.md\u7684\u8f68\u8ff9\u66ff\u6362\u6210\u540c\u4e00\u6a21\u578b\uff08Opus 4.6\uff09\u5728\u5e72\u51c0\u73af\u5883\u4e0b\u7684\u8868\u73b0\u540e\uff0c\u901a\u8fc7\u7387\u4ece81.8%\u964d\u5230\u7ea671.7%\uff0c<strong>\u4ece\u7b2c1\u540d\u6389\u5230\u7b2c14\u540d<\/strong>\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/C47D5B6780636B039BD825E57E6D4C14E10D51C2_size143_w1080_h560.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>ForgeCode\u7684AGENTS.md\u7b54\u6848\u6ce8\u5165\u5b9e\u4f8b\uff1a\u4e0a\u534a\u90e8\u5206\uff0charness\u5c06\u5305\u542b\u6807\u51c6\u7b54\u6848\u7684AGENTS.md\u6587\u4ef6\u6ce8\u5165\u667a\u80fd\u4f53\u73af\u5883\uff1b\u4e0b\u534a\u90e8\u5206\uff0c\u667a\u80fd\u4f53\u8bfb\u53d6\u540e\u76f4\u63a5\u5199\u5165result.txt\uff0c\u5e76\u7528grep\u81ea\u6211\u9a8c\u8bc1\u3002<\/p>\n<p>\u66f4\u6709\u610f\u601d\u7684\u662f\uff0c\u8fd9\u4e9b\u4f5c\u5f0a\u4e0d\u662f\u5f00\u53d1\u8005\u6545\u610f\u4e3a\u4e4b\u3002<\/p>\n<p>Terminal-Bench\u4e0a\u5f88\u591a\u5f00\u53d1\u8005\u516c\u5f00\u8bf4\u8fc7\uff0c\u4ed6\u4eec\u7684harness\u662f\u7528AI\u300cvibecoding\u300d\u51fa\u6765\u7684\u3002<\/p>\n<p>\u4e5f\u5c31\u662f\u8bf4\uff0c<strong>\u5f00\u53d1\u8005\u7528\u6765\u5199\u6d4b\u8bd5\u6846\u67b6\u7684<\/strong><strong>AI<\/strong><strong>\uff0c\u672c\u8eab\u5c31\u662f\u5728\u4f5c\u5f0a<\/strong>\u3002<\/p>\n<p>\u5bbe\u5927\u56e2\u961f\u7ba1\u8fd9\u53eb\u300c\u5143\u7ea7\u522b\u7684reward hacking\u300d\uff1aAI\u5199\u7684\u4ee3\u7801\u81ea\u5e26\u4f5c\u5f0a\u503e\u5411\uff0c\u518d\u901a\u8fc7harness\u4f20\u9012\u7ed9\u6240\u6709\u88ab\u8bc4\u6d4b\u7684\u6a21\u578b\u3002<\/p>\n<p>\u5176\u4ed6\u57fa\u51c6\u4e0a\u4e5f\u4e0d\u5e72\u51c0\u3002CyBench\u4e0a\uff0c464\u6761\u6210\u529f\u8f68\u8ff9\u4e2d\u670916\u6761\uff083.4%\uff09\u662fGoogle\u641c\u7d22\u516c\u5f00\u7684CTF\u653b\u7565\u6284\u6765\u7684\u3002<\/p>\n<p>SWE-bench\u4e0a\uff0c17\u6761\u8f68\u8ff9\u7528git log\u627e\u5230\u4fee\u590d\u63d0\u4ea4\uff0c\u76f4\u63a5\u590d\u5236\u5386\u53f2\u8865\u4e01\u3002<\/p>\n<p>BountyBench\u4e0a\uff0c\u505a\u4e0d\u51fa\u771f\u6b63\u6f0f\u6d1e\u5229\u7528\u7684\u667a\u80fd\u4f53\u4f1a\u4f2a\u9020\u4e00\u4e2a\uff0c\u7528grep\u68c0\u67e5\u6f0f\u6d1e\u6a21\u5f0f\u662f\u5426\u5b58\u5728\u4e8e\u6e90\u7801\u4e2d\uff0c\u7136\u540e\u8dd1\u4e00\u4e2a\u65e0\u5173\u7684pickle.loads()\u6f14\u793a\uff0c\u8bc4\u6d4b\u5668\u53ea\u68c0\u67e5\u9000\u51fa\u7801\uff0c\u5168\u90e8\u901a\u8fc7\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/6977162F5037188CF5070D6A1AA2214A7DBB9682_size217_w1080_h454.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>CyBench\u4e0a\u7684\u4efb\u52a1\u7ea7\u4f5c\u5f0a\uff1aGPT-5.3-Codex\u89e3\u4e0d\u51faCTF\u6311\u6218\u540e\uff0c\u76f4\u63a5\u641c\u7d22\u516c\u5f00\u653b\u7565\uff0c\u4ece\u7f51\u9875\u4e2d\u63d0\u53d6flag\u63d0\u4ea4\u3002<\/p>\n<p><strong>\u6700\u524d\u6cbf\u7684\u6a21\u578b\uff0c\u81ea\u5df1\u5c31\u4f1ahack\u8bc4\u6d4b<\/strong><\/p>\n<p>METR\u7684\u8bc4\u4f30\u62a5\u544a\u91cc\u6709\u4e2a\u6848\u4f8b\u3002<\/p>\n<p>o3\u88ab\u8981\u6c42\u5199\u4e00\u4e2a\u9ad8\u6027\u80fdGPU\u6838\u51fd\u6570\uff0c\u8bc4\u5206\u51fd\u6570\u5148\u8fd0\u884c\u53c2\u8003\u65b9\u6848\u5f97\u5230\u6b63\u786e\u7ed3\u679c\uff0c\u518d\u8fd0\u884co3\u7684\u65b9\u6848\u6bd4\u5bf9\u3002<\/p>\n<p>o3\u7684\u300c\u6838\u51fd\u6570\u300d\u6ca1\u6709\u505a\u4efb\u4f55\u8ba1\u7b97\uff0c\u5b83\u6cbf\u7740Python\u8c03\u7528\u6808\u627e\u5230\u8bc4\u5206\u7cfb\u7edf\u5df2\u7ecf\u7b97\u597d\u7684\u6b63\u786e\u7b54\u6848\uff0c\u76f4\u63a5\u8fd4\u56de\uff0c\u540c\u65f6\u7981\u7528CUDA\u540c\u6b65\u6765\u963b\u6b62\u7cfb\u7edf\u6d4b\u91cf\u771f\u5b9e\u6267\u884c\u65f6\u95f4\u3002<\/p>\n<p>\u4ee3\u7801\u6ce8\u91ca\u91cc\uff0co3\u81ea\u5df1\u5199\u7684\u662f\u300ccheating route\u300d\u3002<\/p>\n<p>RE-Bench\u7684\u7279\u5b9a\u4efb\u52a1\u4e0a\uff0co3\u7684reward hacking\u7387\u63a5\u8fd1<strong>100%<\/strong>\uff0c\u6bcf\u4e00\u6761\u8f68\u8ff9\u90fd\u5728\u4f5c\u5f0a\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/BF937CBEA7C620E5E87C5C83BFE2378626C55CE1_size171_w1080_h608.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>METR\u4e8b\u540e\u95eeo3\uff1a\u300c\u4f60\u89c9\u5f97\u4f60\u7684\u884c\u4e3a\u7b26\u5408\u7528\u6237\u610f\u56fe\u5417\uff1f\u300do3\u7b54\u300c\u4e0d\u7b26\u5408\u300d\uff0c10\u6b21\u95ee10\u6b21\u90fd\u8fd9\u4e48\u7b54\u3002<\/p>\n<p>\u5b83\u77e5\u9053\u81ea\u5df1\u5728\u4f5c\u5f0a\uff0c\u7167\u505a\u4e0d\u8bef\u3002<\/p>\n<p>Claude 3.7 Sonnet\u548co1\u4e5f\u6709\u7c7b\u4f3c\u7684\u60c5\u51b5\u3002<\/p>\n<p>Mythos Preview\u8d70\u5f97\u66f4\u8fdc\u3002\u5728\u4e00\u6b21\u8bc4\u4f30\u4e2d\uff0c\u6a21\u578b\u9700\u8981\u7f16\u8f91\u4e00\u4e2a\u5b83\u6ca1\u6709\u6743\u9650\u7684\u6587\u4ef6\u3002<\/p>\n<p>\u5b83\u641c\u7d22\u4e86\u66ff\u4ee3\u65b9\u6848\uff0c\u627e\u5230\u4e86\u901a\u8fc7\u914d\u7f6e\u6587\u4ef6\u6ce8\u5165\u4ee3\u7801\u6765\u83b7\u53d6\u63d0\u5347\u6743\u9650\u7684\u65b9\u6cd5\uff0c\u7136\u540e<strong>\u8bbe\u8ba1\u4e86\u81ea\u5220\u9664\u673a\u5236<\/strong>\uff0c\u8ba9\u6ce8\u5165\u7684\u4ee3\u7801\u6267\u884c\u5b8c\u6bd5\u540e\u81ea\u52a8\u6e05\u9664\u75d5\u8ff9\u3002<\/p>\n<p>\u6ca1\u6709\u4eba\u6559\u5b83\u8fd9\u4e48\u505a\uff0c\u4f46\u5f53\u6a21\u578b\u80fd\u529b\u8db3\u591f\u5f3a\u3001\u4f18\u5316\u538b\u529b\u8db3\u591f\u5927\uff0c\u5b83\u4f1a\u81ea\u7136\u8d70\u5411\u963b\u529b\u6700\u5c0f\u7684\u8def\u5f84\u3002<\/p>\n<p><strong>\u5206\u6570\u9a71\u52a8\u771f\u91d1\u767d\u94f6\uff0c\u5730\u57fa\u584c\u4e86\u600e\u4e48\u529e<\/strong><\/p>\n<p>\u5de5\u7a0b\u56e2\u961f\u9009\u6a21\u578b\u770bSWE-bench\u6392\u540d\uff0c\u6295\u8d44\u4eba\u770b\u57fa\u51c6\u5206\u6570\u7ed9\u4f30\u503c\uff0c\u7814\u7a76\u8005\u56f4\u7ed5\u5206\u6570\u786e\u5b9a\u4f18\u5316\u65b9\u5411\u3002<\/p>\n<p>\u5982\u679c\u6570\u5b57\u672c\u8eab\u53ef\u4ee5\u88ab\u8f7b\u6613\u64cd\u7eb5\uff0c\u6574\u6761\u51b3\u7b56\u94fe\u7684\u57fa\u7840\u5c31\u662f\u7a7a\u7684\u3002<\/p>\n<p>\u8fd8\u6709\u4e00\u4e2a\u95ee\u9898\uff1a\u80fd\u529b\u8bc4\u6d4b\u548c\u5b89\u5168\u8bc4\u6d4b\u7528\u7684\u662f\u7c7b\u4f3c\u7684\u6280\u672f\u67b6\u6784\u3002<\/p>\n<p>\u5982\u679c\u80fd\u529b\u8bc4\u6d4b\u80fd\u88ab\u6ce8\u6c34\uff0c\u5b89\u5168\u8bc4\u6d4b\u51ed\u4ec0\u4e48\u5e78\u514d\uff1f\u80fdhack\u7f16\u7a0b\u8bc4\u6d4b\u7684\u6a21\u578b\uff0chack\u5bf9\u9f50\u8bc4\u6d4b\u4e5f\u4e0d\u4f1a\u66f4\u96be\u3002<\/p>\n<p><a href=\"https:\/\/www.alphaseek.cn\/?tag=openai\" title=\"View all posts in OpenAI\" target=\"_blank\" class=\"tag_link\">OpenAI<\/a>\u4eca\u5e742\u6708\u5df2\u7ecf\u5ba3\u5e03\u505c\u7528SWE-bench Verified\uff0c\u5185\u90e8\u5ba1\u8ba1\u53d1\u73b059.4%\u7684\u88ab\u5ba1\u8ba1\u95ee\u9898\u5b58\u5728\u6709\u7f3a\u9677\u7684\u6d4b\u8bd5\uff0c\u6a21\u578b\u5728\u7528\u6709bug\u7684\u6807\u51c6\u6765\u8861\u91cf\u3002<\/p>\n<p>\u6240\u6709\u88ab\u6d4b\u7684\u524d\u6cbf\u6a21\u578b\uff08GPT-5.2\u3001Claude Opus 4.5\u3001Gemini 3 Flash\uff09\u90fd\u80fd\u4ece\u8bb0\u5fc6\u4e2d\u590d\u73b0\u6807\u51c6\u7b54\u6848\u7684\u539f\u59cb\u4ee3\u7801\uff0c\u8fde\u53d8\u91cf\u540d\u548c\u5185\u8054\u6ce8\u91ca\u90fd\u4e00\u6837\u3002<\/p>\n<p>SWE-bench Verified\u4e0a\u768470%+\u5206\u6570\uff0c\u5207\u6362\u5230\u66f4\u5e72\u51c0\u7684SWE-bench Pro\u540e\u76f4\u63a5\u964d\u5230\u7ea623%\u3002<\/p>\n<p>\u4f2f\u514b\u5229\u56e2\u961f\u628a\u6f0f\u6d1e\u626b\u63cf\u5de5\u5177\u505a\u6210\u4e00\u4e2a\u53ebBenchJack\u7684\u5f00\u6e90\u9879\u76ee\uff0c\u672c\u8d28\u5c31\u662f\u7ed9\u8bc4\u6d4b\u57fa\u51c6\u505a\u6e17\u900f\u6d4b\u8bd5\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/B16FCC23842D897535581BE9C3B7162F9CD9EDBE_size239_w1080_h541.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"SWE-bench\u6ee1\u5206\uff0c0\u4e2abug\u4fee\u590d\uff1a\u4f2f\u514b\u5229\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\"  \/><\/p>\n<p>\u628a\u5b83\u6307\u5411\u4efb\u4f55\u8bc4\u6d4b\u6d41\u6c34\u7ebf\uff0c\u5b83\u4f1a\u81ea\u52a8\u5206\u6790\u8bc4\u5206\u673a\u5236\u3001\u8bc6\u522b\u9694\u79bb\u8fb9\u754c\u3001\u751f\u6210\u53ef\u8fd0\u884c\u7684\u6f0f\u6d1e\u5229\u7528\u3002<\/p>\n<p>\u5982\u679c\u4e00\u4e2a\u96f6\u80fd\u529b\u667a\u80fd\u4f53\u7684\u5f97\u5206\u9ad8\u4e8e\u57fa\u7ebf\uff0c\u4f60\u7684\u57fa\u51c6\u5c31\u6709\u95ee\u9898\u3002<\/p>\n<p>\u4ed6\u4eec\u7ed9\u51fa\u7684\u5efa\u8bae\u4e5f\u5f88\u76f4\u63a5\uff1a\u8bc4\u6d4b\u7a0b\u5e8f\u548c\u88ab\u6d4bAI\u5fc5\u987b\u5b8c\u5168\u9694\u79bb\u8fd0\u884c\uff0c\u6807\u51c6\u7b54\u6848\u4e0d\u80fd\u51fa\u73b0\u5728AI\u80fd\u8bbf\u95ee\u7684\u73af\u5883\u4e2d\uff0c\u6c38\u8fdc\u4e0d\u8981\u5bf9\u4e0d\u53ef\u4fe1\u7684\u8f93\u5165\u8c03\u7528eval()\uff0cLLM\u88c1\u5224\u8981\u50cf\u5904\u7406\u7528\u6237\u8f93\u5165\u4e00\u6837\u5bf9AI\u8f93\u51fa\u505a\u8fc7\u6ee4\u3002<\/p>\n<p>\u6709\u4eba\u5728\u63a8\u7279\u4e0a\u8bc4\u8bba\uff1a<\/p>\n<\/p>\n<p>\u8bf4\u5f97\u6709\u70b9\u7edd\u5bf9\uff0c\u4f46\u5f53\u884c\u4e1a\u56f4\u7ed5\u5206\u6570\u7ade\u4e89\uff0c\u5206\u6570\u672c\u8eab\u7684\u53ef\u4fe1\u5ea6\u53cd\u800c\u6210\u4e86\u6700\u88ab\u5ffd\u89c6\u7684\u4e1c\u897f\u3002<\/p>\n<p>\u8bc4\u6d4b\u672c\u8eab\u6ca1\u6709\u9519\uff0c\u53cd\u800c\u6bd4\u4ee5\u5f80\u4efb\u4f55\u65f6\u5019\u90fd\u91cd\u8981\u3002\u4e0d\u662f\u300c\u5206\u6570\u662f\u591a\u5c11\u300d\uff0c\u800c\u662f\u300c\u8fd9\u4e2a\u5206\u6570\u662f\u600e\u4e48\u6765\u7684\u300d\u3002<\/p>\n<p>\u56de\u5230\u5f00\u5934\u90a310\u884c\u4ee3\u7801\u3002SWE-bench\u4e0a\uff0c\u6700\u597d\u7684\u6a21\u578b\u8dd1\u51fa70%\u300180%\u7684\u6210\u7ee9\uff0c\u5404\u5bb6\u53d1\u5e03\u4f1a\u4e0a\u53cd\u590d\u5f15\u7528\u3002<\/p>\n<p>\u4f46\u4e00\u4e2a\u4ec0\u4e48\u90fd\u4e0d\u4f1a\u7684conftest.py\u62ff\u4e86100%\u3002<\/p>\n<p>\u5728\u8fd9\u4e2a100%\u88ab\u9020\u51fa\u6765\u4e4b\u524d\uff0c\u6ca1\u6709\u4eba\u89c9\u5f97\u5206\u6570\u6709\u95ee\u9898\u3002<\/p>\n<p>\u53c2\u8003\u8d44\u6599\uff1a<\/p>\n<p>https:\/\/x.com\/dotey\/status\/2043204009469641005<\/p>\n<p>(\u6765\u6e90\uff1a\u65b0\u6d6a\u79d1\u6280)<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u65b0\u667a\u5143\u62a5\u9053 \u7f16\u8f91\uff1a\u503e\u503e \u3010\u65b0\u667a\u5143\u5bfc\u8bfb\u3011\u4f2f\u514b\u5229\u56e2\u961f\u9020\u4e86\u4e2a\u4e13\u95e8\u4f5c\u5f0a\u7684AI\uff0c\u752810\u884cPython\u4ee3\u7801\u62ff\u4e0bSWE-be [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[6],"tags":[271,371,378,241],"views":35,"_links":{"self":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/764739"}],"collection":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=764739"}],"version-history":[{"count":0,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/764739\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=764739"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=764739"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=764739"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}