{"id":728,"date":"2026-08-10T17:13:50","date_gmt":"2026-08-10T08:13:50","guid":{"rendered":"https:\/\/www.wetch.co.jp\/ai\/?p=728"},"modified":"2026-08-10T17:13:50","modified_gmt":"2026-08-10T08:13:50","slug":"%e6%ba%80%e7%82%b9%e3%81%aa%e3%81%ae%e3%81%ab%e4%bd%bf%e3%81%88%e3%81%aa%e3%81%84%e2%80%95%e2%80%952026%e5%b9%b4%e3%80%81ai%e3%83%99%e3%83%b3%e3%83%81%e3%83%9e%e3%83%bc%e3%82%af%e3%81%8c%e4%bf%a1","status":"publish","type":"post","link":"https:\/\/www.wetch.co.jp\/ai\/2026\/08\/10\/%e6%ba%80%e7%82%b9%e3%81%aa%e3%81%ae%e3%81%ab%e4%bd%bf%e3%81%88%e3%81%aa%e3%81%84%e2%80%95%e2%80%952026%e5%b9%b4%e3%80%81ai%e3%83%99%e3%83%b3%e3%83%81%e3%83%9e%e3%83%bc%e3%82%af%e3%81%8c%e4%bf%a1\/","title":{"rendered":"\u6e80\u70b9\u306a\u306e\u306b\u4f7f\u3048\u306a\u3044\u2015\u20152026\u5e74\u3001AI\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u304c\u4fe1\u7528\u3092\u5931\u3063\u3066\u3044\u304f\u7406\u7531"},"content":{"rendered":"<p><em>\u30ea\u30fc\u30c0\u30fc\u30dc\u30fc\u30c9\u306e\u9996\u4f4d\u304c\u5165\u308c\u66ff\u308f\u308b\u305f\u3073\u306b\u30cb\u30e5\u30fc\u30b9\u306b\u306a\u308b\u6642\u4ee3\u306f\u3001\u5b9f\u306f\u9759\u304b\u306b\u7d42\u308f\u308a\u3064\u3064\u3042\u308b\u30022026\u5e74\u306b\u5165\u308a\u3001\u516c\u958b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u9ad8\u30b9\u30b3\u30a2\u304c\u5b9f\u969b\u306e\u6027\u80fd\u3092\u610f\u5473\u3057\u306a\u3044\u3068\u3044\u3046\u6307\u6458\u304c\u76f8\u6b21\u304e\u3001<a href=\"https:\/\/openai.com\/\" target=\"_blank\" rel=\"noopener\">OpenAI<\/a>\u81ea\u8eab\u304c\u770b\u677f\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u4e00\u3064\u3067\u63a1\u70b9\u3092\u3084\u3081\u308b\u4e8b\u614b\u306b\u307e\u3067\u767a\u5c55\u3057\u305f\u3002\u300c\u6e2c\u308b\u307b\u3069\u58ca\u308c\u308b\u300d\u3068\u3044\u3046\u9006\u8aac\u304cAI\u696d\u754c\u306e\u5171\u901a\u8a8d\u8b58\u306b\u306a\u308a\u3064\u3064\u3042\u308b\u3001\u305d\u306e\u5185\u5e55\u3092\u8ffd\u3046\u3002<\/em><\/p>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.wetch.co.jp\/ai\/wp-content\/uploads\/2026\/06\/ill-benchmark.png\" alt=\"AI\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u4fe1\u983c\u6027\u4f4e\u4e0b\u3092\u8868\u3059\u30a4\u30e9\u30b9\u30c8\"\/><\/figure>\n<h2>\u300c\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u5927\u5d29\u58ca\u300d\u3068\u547c\u3070\u308c\u308b\u73fe\u8c61<\/h2>\n<p>\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u80fd\u529b\u306e\u6307\u6a19\u3068\u3057\u3066\u5e83\u304f\u5f15\u7528\u3055\u308c\u3066\u304d\u305f<a href=\"https:\/\/www.swebench.com\/\" target=\"_blank\" rel=\"noopener\">SWE-bench<\/a>\u3002\u305d\u306e\u4ee3\u8868\u7684\u306a\u6d3e\u751f\u7248\u3067\u3042\u308bSWE-bench Verified\u3067\u306f\u3001\u30c8\u30c3\u30d7\u30af\u30e9\u30b9\u306e\u30e2\u30c7\u30eb\u306e\u591a\u304f\u304c8\u5272\u524d\u5f8c\u306e\u30b9\u30b3\u30a2\u3092\u8a18\u9332\u3057\u3066\u304d\u305f\u3002\u3068\u3053\u308d\u304c\u3001\u6c5a\u67d3\u8010\u6027\u3092\u9ad8\u3081\u305f\u65b0\u7248\u300cSWE-bench Pro\u300d\u3067\u540c\u3058\u30e2\u30c7\u30eb\u3092\u6e2c\u308b\u3068\u3001\u6b63\u7b54\u7387\u306f\u304a\u3088\u305d2\u5272\u7a0b\u5ea6\u307e\u3067\u843d\u3061\u8fbc\u3080\u3068\u3044\u3046\u5206\u6790\u304c\u5831\u3058\u3089\u308c\u3066\u3044\u308b\u3002<\/p>\n<p>\u539f\u56e0\u3068\u3057\u3066\u6307\u6458\u3055\u308c\u3066\u3044\u308b\u306e\u304c\u300c\u6697\u8a18\u300d\u3060\u3002\u5b66\u7fd2\u30c7\u30fc\u30bf\u306b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u554f\u984c\u3068\u6b63\u89e3\u304c\u305d\u306e\u307e\u307e\u6df7\u5165\u3057\u3001\u30e2\u30c7\u30eb\u304c\u554f\u984c\u6587\u306e\u7d9a\u304d\u3092\u4e00\u5b57\u4e00\u53e5\u6b63\u78ba\u306b\u88dc\u5b8c\u3067\u304d\u3066\u3057\u307e\u3046\u2014\u2014\u3068\u3044\u3046\u691c\u8a3c\u7d50\u679c\u3092\u3001AI\u5411\u3051\u5b66\u7fd2\u30c7\u30fc\u30bf\u4f01\u696d\u306e<a href=\"https:\/\/surgehq.ai\/\" target=\"_blank\" rel=\"noopener\">Surge AI<\/a>\u304c\u5831\u544a\u3057\u3066\u3044\u308b\u3002\u540c\u793e\u30d6\u30ed\u30b0\u300c<a href=\"https:\/\/surgehq.ai\/blog\/benchmarks-are-broken\" target=\"_blank\" rel=\"noopener\">Benchmarks are broken<\/a>\u300d\u3067\u306f\u3001\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u5404\u793e\u304c\u30e2\u30c7\u30eb\u30ab\u30fc\u30c9\u3067\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u30b9\u30b3\u30a2\u3092\u5f15\u7528\u3057\u306a\u304c\u3089\u3001\u3053\u3046\u3057\u305f\u6c5a\u67d3\u306e\u53ef\u80fd\u6027\u306b\u307b\u3068\u3093\u3069\u89e6\u308c\u3066\u3044\u306a\u3044\u5b9f\u614b\u3092\u6307\u6458\u3057\u3066\u3044\u308b\u3002<\/p>\n<p>\u3053\u308c\u306f\u7d4c\u6e08\u5b66\u30fb\u7d71\u8a08\u5b66\u3067\u3044\u3046\u300c\u30b0\u30c3\u30c9\u30cf\u30fc\u30c8\u306e\u6cd5\u5247\u300d\u2014\u2014\u300c\u6e2c\u5b9a\u5bfe\u8c61\u304c\u76ee\u6a19\u306b\u306a\u3063\u305f\u77ac\u9593\u3001\u305d\u306e\u6e2c\u5b9a\u306f\u610f\u5473\u3092\u5931\u3046\u300d\u2014\u2014\u304cAI\u8a55\u4fa1\u306e\u4e16\u754c\u3067\u305d\u306e\u307e\u307e\u8d77\u304d\u3066\u3044\u308b\u3001\u3068\u6349\u3048\u308b\u3068\u5206\u304b\u308a\u3084\u3059\u3044\u3002<\/p>\n<h2>OpenAI\u304c\u81ea\u3089\u300c\u63a1\u70b9\u3092\u3084\u3081\u305f\u300d\u30d9\u30f3\u30c1\u30de\u30fc\u30af<\/h2>\n<p>\u8c61\u5fb4\u7684\u3060\u3063\u305f\u306e\u304c\u3001OpenAI\u304cSWE-bench Verified\u306e\u30b9\u30b3\u30a2\u5831\u544a\u3092\u53d6\u308a\u3084\u3081\u305f\u3053\u3068\u3060\u3002\u65e5\u672c\u8a9e\u3067\u3053\u306e\u7d4c\u7def\u3092\u4e01\u5be7\u306b\u307e\u3068\u3081\u305f<a href=\"https:\/\/note.com\/takutakuai\/n\/n9a991de0d774\" target=\"_blank\" rel=\"noopener\">\u89e3\u8aac\u8a18\u4e8b<\/a>\u306b\u3088\u308c\u3070\u3001OpenAI\u306f\u5931\u6557\u30bf\u30b9\u30af\u3092\u518d\u691c\u8a3c\u3057\u305f\u3068\u3053\u308d\u3001\u305d\u306e\u7d046\u5272\u3067\u30c6\u30b9\u30c8\u8a2d\u8a08\u305d\u306e\u3082\u306e\u306b\u6b20\u9665\u304c\u3042\u3063\u305f\u3068\u3044\u3046\u3002\u5177\u4f53\u7684\u306b\u306f\u3001\u30c6\u30b9\u30c8\u6761\u4ef6\u304c\u904e\u5ea6\u306b\u53b3\u683c\u3059\u304e\u308b\u30b1\u30fc\u30b9\u304c\u7d0435%\u3001\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u5316\u3055\u308c\u3066\u3044\u306a\u3044\u6319\u52d5\u3092\u6697\u9ed9\u306e\u524d\u63d0\u306b\u3057\u3066\u3044\u305f\u30b1\u30fc\u30b9\u304c\u7d0419%\u3092\u5360\u3081\u305f\u3002\u3064\u307e\u308a\u300c\u30e2\u30c7\u30eb\u304c\u9593\u9055\u3048\u305f\u300d\u306e\u3067\u306f\u306a\u304f\u300c\u63a1\u70b9\u57fa\u6e96\u304c\u305d\u3082\u305d\u3082\u58ca\u308c\u3066\u3044\u305f\u300d\u30bf\u30b9\u30af\u304c\u591a\u6570\u3092\u5360\u3081\u3066\u3044\u305f\u3053\u3068\u306b\u306a\u308b\u3002<\/p>\n<p>OpenAI\u306f\u3053\u306e\u5206\u6790\u3092\u8e0f\u307e\u3048\u3001\u300c\u30b9\u30b3\u30a2\u306e\u5411\u4e0a\u304c\u3082\u306f\u3084\u5b9f\u52d9\u80fd\u529b\u306e\u5411\u4e0a\u3092\u610f\u5473\u305b\u305a\u3001\u3080\u3057\u308d\u7279\u5b9a\u306e\u30c6\u30b9\u30c8\u5f62\u5f0f\u306b\u3069\u308c\u3060\u3051\u6700\u9069\u5316\u3057\u305f\u304b\u3092\u53cd\u6620\u3057\u3066\u3044\u308b\u306b\u904e\u304e\u306a\u3044\u300d\u3068\u3044\u3046\u8da3\u65e8\u306e\u898b\u89e3\u3092\u793a\u3057\u3066\u3044\u308b\u3002\u770b\u677f\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u4e00\u3064\u3092\u958b\u767a\u5143\u81ea\u8eab\u304c\u300c\u3082\u3046\u4fe1\u7528\u3067\u304d\u306a\u3044\u300d\u3068\u516c\u8a00\u3057\u305f\u683c\u597d\u3060\u3002<\/p>\n<h2>\u300c\u30d9\u30f3\u30c1\u30de\u30af\u30b7\u30f3\u30b0\u300d\u3068\u3044\u3046\u30b9\u30e9\u30f3\u30b0<\/h2>\n<p>\u3053\u306e\u72b6\u6cc1\u3092\u6307\u3057\u3066\u696d\u754c\u3067\u4f7f\u308f\u308c\u59cb\u3081\u305f\u306e\u304c<strong>\u30d9\u30f3\u30c1\u30de\u30af\u30b7\u30f3\u30b0(benchmaxxing)<\/strong>\u3068\u3044\u3046\u8a00\u8449\u3060\u3002\u516c\u958b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u9ad8\u5f97\u70b9\u3092\u53d6\u308b\u3053\u3068\u3060\u3051\u3092\u76ee\u7684\u306b\u30e2\u30c7\u30eb\u3092\u6700\u9069\u5316\u3059\u308b\u884c\u70ba\u3092\u6307\u3059\u3002<a href=\"https:\/\/arxiv.org\/html\/2607.22368v1\" target=\"_blank\" rel=\"noopener\">arXiv\u306e\u8ad6\u6587\u300cDo Agent Benchmarks Measure Capability?\u300d<\/a>\u3067\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u578b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306b\u304a\u3044\u3066\u30e2\u30c7\u30eb\u304c\u63a1\u70b9\u7528\u306e\u30e1\u30bf\u30c7\u30fc\u30bf\u3092\u76f4\u63a5\u63a2\u7d22\u3057\u305f\u308a\u3001\u96a0\u3055\u308c\u3066\u3044\u308b\u306f\u305a\u306e\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u3092\u518d\u69cb\u6210\u3057\u305f\u308a\u3059\u308b\u300c\u30ea\u30ef\u30fc\u30c9\u30cf\u30c3\u30ad\u30f3\u30b0\u300d\u306e\u5b9f\u4f8b\u304c\u8907\u6570\u5831\u544a\u3055\u308c\u3066\u3044\u308b\u3002\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3092\u4f5c\u308b\u5074\u3068\u3001\u305d\u308c\u3092\u653b\u7565\u3059\u308b\u5074\u306e\u3044\u305f\u3061\u3054\u3063\u3053\u304c\u3001\u30c6\u30b9\u30c8\u81ea\u4f53\u306e\u59a5\u5f53\u6027\u3092\u6398\u308a\u5d29\u3057\u3066\u3044\u308b\u69cb\u56f3\u3060\u3002<\/p>\n<p>\u8457\u540d\u306a\u7814\u7a76\u8005Andrej Karpathy\u6c0f\u304c\u3001\u516c\u958b\u30ea\u30fc\u30c0\u30fc\u30dc\u30fc\u30c9\u300cLMArena\u300d\u306e\u4e0a\u4f4d\u30e2\u30c7\u30eb\u3068\u81ea\u8eab\u306e\u4f7f\u7528\u611f\u304c\u4e00\u81f4\u3057\u306a\u3044\u3068\u6307\u6458\u3057\u305f\u3053\u3068\u3082\u8a71\u984c\u306b\u306a\u3063\u305f\u3002\u6570\u5b57\u306e\u4e0a\u306e\u4e00\u4f4d\u3068\u3001\u5b9f\u969b\u306b\u4f7f\u3063\u3066\u300c\u826f\u3044\u300d\u3068\u611f\u3058\u308b\u30e2\u30c7\u30eb\u304c\u4e00\u81f4\u3057\u306a\u3044\u2014\u2014\u3053\u306e\u4e56\u96e2\u3053\u305d\u304c\u30012026\u5e74\u306eAI\u8a55\u4fa1\u304c\u76f4\u9762\u3057\u3066\u3044\u308b\u6838\u5fc3\u7684\u306a\u554f\u984c\u3068\u3044\u3048\u308b\u3002<\/p>\n<table>\n<thead>\n<tr>\n<th>\u6307\u6a19<\/th>\n<th>\u3053\u308c\u307e\u3067\u306e\u59ff<\/th>\n<th>2026\u5e74\u306b\u9732\u5448\u3057\u305f\u554f\u984c<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>SWE-bench Verified<\/td>\n<td>\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u80fd\u529b\u306e\u4ee3\u8868\u7684\u306a\u6307\u6a19<\/td>\n<td>\u5b66\u7fd2\u30c7\u30fc\u30bf\u6c5a\u67d3\u306b\u3088\u308a\u6697\u8a18\u3067\u9ad8\u5f97\u70b9\u3001OpenAI\u304c\u63a1\u70b9\u3092\u53d6\u308a\u3084\u3081<\/td>\n<\/tr>\n<tr>\n<td>MMLU \/ HumanEval<\/td>\n<td>\u6c4e\u7528\u77e5\u8b58\u30fb\u30b3\u30fc\u30c9\u751f\u6210\u306e\u5b9a\u756a\u30d9\u30f3\u30c1\u30de\u30fc\u30af<\/td>\n<td>\u30c8\u30c3\u30d7\u30e2\u30c7\u30eb\u304c\u8ed2\u4e26\u307f9\u5272\u8d85\u3067\u300c\u98fd\u548c\u300d\u3057\u5dee\u304c\u3064\u304b\u306a\u304f\u306a\u3063\u305f<\/td>\n<\/tr>\n<tr>\n<td>LMArena(\u65e7Chatbot Arena)<\/td>\n<td>\u4eba\u9593\u6295\u7968\u306b\u3088\u308b\u30e2\u30c7\u30eb\u6bd4\u8f03<\/td>\n<td>\u9806\u4f4d\u3068\u5b9f\u4f7f\u7528\u611f\u306e\u4e56\u96e2\u304c\u6307\u6458\u3055\u308c\u308b<\/td>\n<\/tr>\n<tr>\n<td>SWE-bench Pro \u306a\u3069\u8010\u6c5a\u67d3\u7248<\/td>\n<td>\u65b0\u8a2d\u30fb\u5f37\u5316\u4e2d<\/td>\n<td>\u30b9\u30b3\u30a2\u304c\u5927\u5e45\u306b\u4e0b\u304c\u308a\u300c\u672c\u5f53\u306e\u5b9f\u529b\u300d\u306b\u8fd1\u3044\u6570\u5b57\u304c\u51fa\u308b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>\u6570\u5b57\u3092\u9d5c\u5451\u307f\u306b\u3057\u306a\u3044\u305f\u3081\u306e3\u3064\u306e\u8996\u70b9<\/h2>\n<ul>\n<li><strong>\u81ea\u5df1\u7533\u544a\u304b\u7b2c\u4e09\u8005\u691c\u8a3c\u304b\u3092\u533a\u5225\u3059\u308b\u3002<\/strong>\u30e2\u30c7\u30eb\u30ab\u30fc\u30c9\u306b\u66f8\u304b\u308c\u305f\u30b9\u30b3\u30a2\u306f\u30d9\u30f3\u30c0\u30fc\u81ea\u8eab\u306e\u7533\u544a\u3067\u3042\u308b\u3053\u3068\u304c\u591a\u3044\u3002<a href=\"https:\/\/epoch.ai\/\" target=\"_blank\" rel=\"noopener\">Epoch AI<\/a>\u3084<a href=\"https:\/\/artificialanalysis.ai\/\" target=\"_blank\" rel=\"noopener\">Artificial Analysis<\/a>\u306e\u3088\u3046\u306b\u3001\u72ec\u81ea\u30cf\u30fc\u30cd\u30b9\u3067\u8ffd\u8a66\u3059\u308b\u7b2c\u4e09\u8005\u6a5f\u95a2\u306e\u6570\u5b57\u3068\u7a81\u304d\u5408\u308f\u305b\u308b\u3068\u3001\u5b9f\u614b\u304c\u305a\u308c\u3066\u3044\u308b\u30b1\u30fc\u30b9\u304c\u5c11\u306a\u304f\u306a\u3044\u3002<\/li>\n<li><strong>\u30b9\u30b3\u30a2\u306e\u300c\u5e45\u300d\u3092\u898b\u308b\u3002<\/strong>AI\u306e\u81ea\u5f8b\u4f5c\u696d\u80fd\u529b\u3092\u8a08\u6e2c\u3059\u308b<a href=\"https:\/\/metr.org\/\" target=\"_blank\" rel=\"noopener\">METR<\/a>\u306f\u3001\u30e2\u30c7\u30eb\u304c50%\u306e\u78ba\u7387\u3067\u3053\u306a\u305b\u308b\u30bf\u30b9\u30af\u306e\u9577\u3055(\u30bf\u30b9\u30af\u30fb\u30b3\u30f3\u30d7\u30ea\u30fc\u30b7\u30e7\u30f3\u30fb\u30bf\u30a4\u30e0\u30db\u30e9\u30a4\u30ba\u30f3)\u3092\u6307\u6a19\u306b\u3057\u3066\u3044\u308b\u304c\u3001\u305d\u306e95%\u4fe1\u983c\u533a\u9593\u306f6\u6642\u9593\u304b\u308998\u6642\u9593\u307e\u3067\u4e00\u6841\u4ee5\u4e0a\u306e\u5e45\u304c\u3042\u308b\u3002\u5358\u4e00\u306e\u6570\u5b57\u3088\u308a\u3001\u3070\u3089\u3064\u304d\u306e\u5927\u304d\u3055\u81ea\u4f53\u304c\u5b9f\u60c5\u3092\u8a9e\u3063\u3066\u3044\u308b\u3002<\/li>\n<li><strong>\u30b3\u30b9\u30c8\u3068\u518d\u73fe\u6027\u3067\u898b\u308b\u3002<\/strong>\u4fe1\u983c\u3067\u304d\u308b\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5411\u3051\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u69cb\u7bc9\u306b\u306f\u6570\u5104\u5186\u898f\u6a21\u306e\u521d\u671f\u8cbb\u7528\u3068\u5e74\u9593\u6570\u5343\u4e07\u5186\u898f\u6a21\u306e\u7dad\u6301\u8cbb\u304c\u304b\u304b\u308b\u3068Surge AI\u306f\u8a66\u7b97\u3057\u3066\u3044\u308b\u3002\u5b89\u4fa1\u3067\u7c21\u5358\u306b\u4f5c\u308c\u308b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u307b\u3069\u3001\u653b\u7565\u3055\u308c\u5c3d\u304f\u3057\u3066\u3044\u308b\u53ef\u80fd\u6027\u3092\u7591\u3063\u305f\u307b\u3046\u304c\u3044\u3044\u3002<\/li>\n<\/ul>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.wetch.co.jp\/ai\/wp-content\/uploads\/2026\/08\/exam-hall-canton.jpg\" alt=\"1873\u5e74\u3001\u5e83\u5dde\u306e\u79d1\u6319\u8a66\u9a13\u5834\u306e\u5199\u771f\"\/><figcaption><small>Photo: <a href=\"https:\/\/www.flickr.com\/photos\/73175123@N02\" target=\"_blank\" rel=\"noopener\">ralphrepo<\/a> \/ CC BY 2.0\uff08Openverse\u7d4c\u7531\uff09\u30001873\u5e74\u3001\u5e83\u5dde\u306e\u79d1\u6319(\u5b98\u540f\u767b\u7528\u8a66\u9a13)\u4f1a\u5834\u3002\u300c\u8a66\u9a13\u306e\u70b9\u6570\u3068\u5b9f\u529b\u304c\u4e00\u81f4\u3059\u308b\u3068\u306f\u9650\u3089\u306a\u3044\u300d\u3068\u3044\u3046\u554f\u3044\u306f\u3001AI\u306b\u9650\u3089\u305a\u8a66\u9a13\u5236\u5ea6\u305d\u306e\u3082\u306e\u306b\u3064\u304d\u307e\u3068\u3063\u3066\u304d\u305f\u53e4\u3044\u8ab2\u984c\u3067\u3082\u3042\u308b\u3002<\/small><\/figcaption><\/figure>\n<h2>\u3053\u308c\u304b\u3089 \u2014 \u300c\u5be9\u5224\u300d\u81ea\u4f53\u304c\u8a55\u4fa1\u3055\u308c\u308b\u6642\u4ee3\u3078<\/h2>\n<p>\u516c\u958b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3078\u306e\u4e0d\u4fe1\u304c\u5e83\u304c\u3063\u305f\u7d50\u679c\u3001\u72ec\u7acb\u7cfb\u306e\u8a55\u4fa1\u6a5f\u95a2\u304c\u5b58\u5728\u611f\u3092\u5897\u3057\u3066\u3044\u308b\u3002METR\u3001Epoch AI\u3001Artificial Analysis\u3068\u3044\u3063\u305f\u7b2c\u4e09\u8005\u304c\u3001\u30d9\u30f3\u30c0\u30fc\u306e\u767a\u8868\u3092\u9d5c\u5451\u307f\u306b\u305b\u305a\u81ea\u524d\u306e\u30cf\u30fc\u30cd\u30b9\u3067\u8ffd\u8a66\u3057\u3001\u305d\u306e\u624b\u6cd5\u81ea\u4f53\u3092\u516c\u958b\u3057\u3066\u691c\u8a3c\u53ef\u80fd\u306b\u3059\u308b\u6d41\u308c\u304c\u5f37\u307e\u3063\u3066\u3044\u308b\u3002\u7814\u7a76\u8005\u3082\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u8a2d\u8a08\u3092\u300c\u5236\u5ea6\u8a2d\u8a08\u300d\u306e\u554f\u984c\u3068\u3057\u3066\u6349\u3048\u76f4\u3057\u3066\u304a\u308a\u3001\u6cd5\u5f8b\u5206\u91ce\u5411\u3051AI\u8a55\u4fa1\u3092\u8ad6\u3058\u305f<a href=\"https:\/\/www.pnas.org\/doi\/10.1073\/pnas.2509757122\" target=\"_blank\" rel=\"noopener\">PNAS\u63b2\u8f09\u306e\u8ad6\u6587<\/a>\u306f\u305d\u306e\u4e00\u4f8b\u3060\u3002<\/p>\n<p>\u79c1\u305f\u3061\u30e6\u30fc\u30b6\u30fc\u306b\u3068\u3063\u3066\u306e\u5b9f\u52d9\u7684\u306a\u6559\u8a13\u306f\u30b7\u30f3\u30d7\u30eb\u3060\u3002\u300c\u3069\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u4f55\u4f4d\u304b\u300d\u3088\u308a\u3082\u3001\u300c\u8ab0\u304c\u6e2c\u3063\u305f\u304b\u300d\u300c\u3069\u306e\u304f\u3089\u3044\u306e\u3070\u3089\u3064\u304d\u304c\u3042\u308b\u304b\u300d\u300c\u81ea\u5206\u306e\u696d\u52d9\u306b\u8fd1\u3044\u30bf\u30b9\u30af\u3067\u8a66\u3057\u305f\u304b\u300d\u3092\u898b\u308b\u7fd2\u6163\u306b\u5207\u308a\u66ff\u3048\u308b\u6642\u671f\u306b\u6765\u3066\u3044\u308b\u3002\u30ea\u30fc\u30c0\u30fc\u30dc\u30fc\u30c9\u306e\u4e00\u4f4d\u8868\u793a\u306f\u3001\u3082\u306f\u3084\u54c1\u8cea\u306e\u8a3c\u660e\u66f8\u3067\u306f\u306a\u304f\u3001\u8a71\u534a\u5206\u306b\u805e\u304f\u3079\u304d\u5e83\u544a\u30b3\u30d4\u30fc\u306b\u8fd1\u3065\u304d\u3064\u3064\u3042\u308b\u3002<\/p>\n<hr \/>\n<h2>\u4e3b\u306a\u60c5\u5831\u6e90<\/h2>\n<ul>\n<li><a href=\"https:\/\/surgehq.ai\/blog\/benchmarks-are-broken\" target=\"_blank\" rel=\"noopener\">Surge AI Blog: Benchmarks are broken<\/a><\/li>\n<li><a href=\"https:\/\/note.com\/takutakuai\/n\/n9a991de0d774\" target=\"_blank\" rel=\"noopener\">note: SWE-bench\u306e\u6e80\u70b9\u306f\u8ab0\u306e\u3082\u306e\u304b\u2014\u2014\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u6c5a\u67d3\u6642\u4ee3\u306b\u30c1\u30fc\u30e0\u304c\u3059\u3079\u304d3\u3064\u306e\u3053\u3068<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/html\/2607.22368v1\" target=\"_blank\" rel=\"noopener\">arXiv: Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI<\/a><\/li>\n<li><a href=\"https:\/\/www.pnas.org\/doi\/10.1073\/pnas.2509757122\" target=\"_blank\" rel=\"noopener\">PNAS: There is no free benchmark: An institutional view of legal AI benchmarking<\/a><\/li>\n<li><a href=\"https:\/\/www.swebench.com\/\" target=\"_blank\" rel=\"noopener\">SWE-bench \u516c\u5f0f\u30b5\u30a4\u30c8<\/a><\/li>\n<li><a href=\"https:\/\/metr.org\/\" target=\"_blank\" rel=\"noopener\">METR \u516c\u5f0f\u30b5\u30a4\u30c8<\/a><\/li>\n<li><a href=\"https:\/\/epoch.ai\/\" target=\"_blank\" rel=\"noopener\">Epoch AI \u516c\u5f0f\u30b5\u30a4\u30c8<\/a><\/li>\n<li><a href=\"https:\/\/artificialanalysis.ai\/\" target=\"_blank\" rel=\"noopener\">Artificial Analysis \u516c\u5f0f\u30b5\u30a4\u30c8<\/a><\/li>\n<\/ul>\n<p><small>\u5404\u30c4\u30fc\u30eb\u306f\u63d0\u4f9b\u5143\u306e\u898f\u7d04\u306b\u5f93\u3063\u3066\u3054\u5229\u7528\u304f\u3060\u3055\u3044\u3002\u60c5\u5831\u306f\u5404\u60c5\u5831\u6e90\u306e\u516c\u958b\u6642\u70b9\u306e\u3082\u306e\u3067\u3059\u3002\u53ce\u96c6\u30fb\u8981\u7d04\u306fAI\u3001\u516c\u958b\u5224\u65ad\u306f\u4eba\u3002<\/small><\/p>\n","protected":false},"excerpt":{"rendered":"<p>SWE-bench\u306e\u300c\u4e38\u6697\u8a18\u300d\u767a\u899a\u3001OpenAI\u306b\u3088\u308b\u30b9\u30b3\u30a2\u5831\u544a\u306e\u53d6\u308a\u3084\u3081\u3001\u300c\u30d9\u30f3\u30c1\u30de\u30af\u30b7\u30f3\u30b0\u300d\u3068\u3044\u3046\u65b0\u8a9e\u306e\u767b\u5834\u2014\u20142026\u5e74\u3001\u516c\u958b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u6570\u5b57\u304c\u5b9f\u529b\u306e\u4ee3\u308f\u308a\u306b\u306a\u3089\u306a\u304f\u306a\u3063\u3066\u304d\u305f\u80cc\u666f\u3068\u3001\u305d\u306e\u5148\u306b\u751f\u307e\u308c\u3064\u3064\u3042\u308b\u65b0\u3057\u3044\u8a55\u4fa1\u306e\u304b\u305f\u3061\u3092\u6574\u7406\u3059\u308b\u3002<\/p>\n","protected":false},"author":1,"featured_media":46,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"page_builder":"","footnotes":""},"categories":[15],"tags":[],"class_list":["post-728","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-15"],"_links":{"self":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/728","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/comments?post=728"}],"version-history":[{"count":1,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/728\/revisions"}],"predecessor-version":[{"id":742,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/posts\/728\/revisions\/742"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/media\/46"}],"wp:attachment":[{"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/media?parent=728"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/categories?post=728"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wetch.co.jp\/ai\/wp-json\/wp\/v2\/tags?post=728"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}