
{"id":9149,"date":"2026-06-16T04:33:12","date_gmt":"2026-06-15T20:33:12","guid":{"rendered":"https:\/\/infernews.com\/blog\/mbench\/"},"modified":"2026-06-16T04:34:42","modified_gmt":"2026-06-15T20:34:42","slug":"mbench","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/mbench\/","title":{"rendered":"MBench \u5c08\u770b\u9577\u5f71\u7247\u4e16\u754c\u6a21\u578b\u8a18\u61b6\u529b"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-f7791d216d32.jpg\" alt=\"Teaser\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6642\u4e0d\u5c11\u9577\u5f71\u7247\u8a55\u6e2c\uff0c\u4ecd\u504f\u5411\u55ae\u5e40\u756b\u8cea\u6216\u77ed\u8ddd\u96e2 prompt following\uff1b\u756b\u9762\u4e00\u65e6\u5207\u8d70\u3001\u89d2\u8272\u96e2\u93e1\uff0c\u5f88\u591a\u6a21\u578b\u4fbf\u5bb9\u6613\u5728\u56de\u5230\u540c\u4e00\u60c5\u5883\u6642\u300c\u5931\u61b6\u300d\u3002MBench \u9019\u500b<strong>benchmark<\/strong>\u6b63\u662f\u91dd\u5c0d\u9019\u500b\u76f2\u9ede\u800c\u8a2d\uff0c\u805a\u7126 long-video world models \u7684 memory capability\uff0c\u6aa2\u67e5\u6a21\u578b\u80fd\u5426\u5728\u6642\u9593\u62c9\u9577\u5f8c\u7dad\u6301\u4e00\u81f4\u7684\u4e16\u754c\u72c0\u614b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f5c\u8005\u628a\u554f\u984c\u62c6\u6210\u4e09\u500b\u4e92\u76f8\u7368\u7acb\u4f46\u53c8\u5f7c\u6b64\u95dc\u9023\u7684\u65b9\u5411\uff1aEntity Consistency\u3001Environment Consistency\u3001Causal Consistency\u3002\u9019\u7a2e\u8a2d\u8a08\u6bd4\u7c60\u7d71\u5730\u7d66\u4e00\u500b\u7e3d\u5206\u66f4\u6709\u5206\u6790\u50f9\u503c\uff0c\u56e0\u70ba\u4f60\u80fd\u770b\u6e05\u6a21\u578b\u7a76\u7adf\u662f\u5fd8\u8a18\u89d2\u8272\u5916\u89c0\u3001\u641e\u4e82\u5834\u666f\u7a7a\u9593\uff0c\u9084\u662f\u672a\u80fd\u5ef6\u7e8c\u756b\u9762\u5916\u4ecd\u5728\u767c\u751f\u7684\u7269\u7406\u904e\u7a0b\uff1b\u540c\u6642\u5b83\u518d\u5206\u6210 MBench-A \u8207 MBench-T\uff0c\u5206\u5225\u5c0d\u61c9 action-conditioned world models \u8207 text-segment-conditioned \u9577\u5f71\u7247\u7e8c\u5beb\u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u4f60\u672c\u8eab\u6709\u5f71\u7247\u751f\u6210\u6216\u4e16\u754c\u6a21\u578b\u9805\u76ee\uff0c\u9019\u500b\u5132\u5b58\u5eab\u7684\u7528\u9014\u5f88\u660e\u78ba\uff1a\u5148\u6e96\u5099\u6a21\u578b\u8f38\u51fa\uff0c\u518d\u7528 <code>mbench<\/code> \u9019\u5957 contract-driven\u3001plugin-based CLI \u8dd1\u5b8c\u6574\u8a55\u6e2c\u6d41\u7a0b\u3002\u5132\u5b58\u5eab\u5df2\u63d0\u4f9b 12 \u500b\u5b98\u65b9 metric implementation\uff0c\u4ea6\u6574\u5408 VLM trigger judge\uff0c\u4ee3\u8868\u5b83\u4e0d\u53ea\u662f\u8ad6\u6587\u6982\u5ff5\uff0c\u800c\u662f\u4e00\u5957\u53ef\u843d\u5730\u6bd4\u8f03\u4e0d\u540c\u6a21\u578b\u8868\u73fe\u7684\u8a55\u6e2c\u5de5\u5177\u93c8\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9805\u76ee\u985e\u578b<\/strong>\uff1a\u9019\u662f\u4e00\u500b benchmark\uff0f\u8a55\u6e2c\u5de5\u5177\u93c8\uff0c\u7528\u4f86\u91cf\u5ea6\u9577\u5f71\u7247\u4e16\u754c\u6a21\u578b\u662f\u5426\u5177\u5099\u7a69\u5b9a\u8a18\u61b6\u8207\u6642\u5e8f\u4e00\u81f4\u6027\u3002<\/li>\n\n\n\n<li><strong>\u5275\u65b0\u4f4d\u7f6e<\/strong>\uff1a\u4e0d\u662f\u53ea\u770b\u756b\u9762\u8cea\u7d20\uff0c\u800c\u662f\u628a\u300c\u9577\u6642\u9593\u8a18\u4f4f\u4e16\u754c\u300d\u6b63\u5f0f\u5b9a\u7fa9\u6210\u4e09\u689d capability axes\u3002<\/li>\n\n\n\n<li><strong>\u9069\u5408\u5834\u666f<\/strong>\uff1a\u9577\u5f71\u7247\u751f\u6210\u3001world model \u7814\u7a76\u3001\u6a21\u578b\u6bd4\u8f03\u3001\u5167\u90e8\u9a57\u8b49\u65b0\u7248\u672c\u9000\u6b65\u8207\u5426\u3002<\/li>\n\n\n\n<li><strong>\u53ef\u8b80\u6027\u9ad8<\/strong>\uff1aMBench-A \u8207 MBench-T \u5c07\u4e0d\u540c\u689d\u4ef6\u8a2d\u5b9a\u5206\u958b\uff0c\u8f03\u5bb9\u6613\u77e5\u9053\u6a21\u578b\u5931\u5206\u539f\u56e0\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f9e\u8a55\u8ad6\u89d2\u5ea6\u770b\uff0cMBench \u7684\u50f9\u503c\u5728\u65bc\u5b83\u6279\u8a55\u4e86\u820a\u6709\u56fa\u5b9a\u7bc4\u5f0f\uff1a\u53ea\u734e\u52f5 single-frame quality \u6216 short-horizon prompt following\uff0c\u537b\u672a\u6709\u6aa2\u9a57\u8de8\u93e1\u982d\u3001\u8de8\u6642\u9593\u7684\u6301\u7e8c\u8a18\u61b6\u3002\u82e5\u4f60\u95dc\u5fc3\u7684\u6a21\u578b\u5305\u62ec\u5404\u985e long-video world models\u3001action-conditioned world models\uff0c\u4ee5\u53ca text continuation \u985e\u5f71\u7247\u6a21\u578b\uff0c\u9019\u500b\u9805\u76ee\u5f88\u503c\u5f97\u7d0d\u5165\u6e2c\u8a66\u6d41\u7a0b\uff1b\u4e0d\u904e\u76ee\u524d\u63d0\u4f9b\u7684\u8cc7\u6599\u4ee5 benchmark \u8207\u8a55\u6e2c\u6846\u67b6\u70ba\u4e3b\uff0c\u662f\u5426\u80fd\u5168\u9762\u4ee3\u8868\u6240\u6709\u771f\u5be6\u5275\u4f5c\u5834\u666f\uff0c\u4ecd\u8981\u914d\u5408\u4f60\u81ea\u5df1\u7684\u751f\u6210\u4efb\u52d9\u4e00\u4f75\u89c0\u5bdf\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/study-overflow\/MBench\" rel=\"noopener noreferrer\">https:\/\/github.com\/study-overflow\/MBench<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2606.00793\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2606.00793<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>MBench\u4e0d\u662f\u770b\u55ae\u5e40\u975a\u5514\u975a\uff0c\u800c\u662f\u8ffd\u554f\u6a21\u578b\u80fd\u5426\u8a18\u4f4f\u96e2\u93e1\u5f8c\u518d\u51fa\u73fe\u7684\u4eba\u8207\u5834\u666f\u3002\u5b83\u628a\u9577\u5f71\u7247\u8a18\u61b6\u62c6\u6210\u4e09\u689d\u80fd\u529b\u8ef8\uff0c\u4ee4\u8a55\u6e2c\u66f4\u8cbc\u8fd1\u751f\u6210\u9023\u7e8c\u5f71\u7247\u7684\u96e3\u984c\u3002<\/p>\n","protected":false},"author":8,"featured_media":9148,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,31,38,167,150,76,141,186,196,197],"tags":[],"class_list":["post-9149","post","type-post","status-publish","format-standard","hentry","category-133","category-video","category-38","category-win","category-150","category-76","category-141","category-186","category-196","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9149","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9149"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9149\/revisions"}],"predecessor-version":[{"id":9151,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9149\/revisions\/9151"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9148"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9149"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9149"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9149"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}