{"id":6029,"date":"2024-09-13T19:41:21","date_gmt":"2024-09-13T11:41:21","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=6029"},"modified":"2024-11-22T07:49:58","modified_gmt":"2024-11-21T23:49:58","slug":"openai-o1-mini-dabiao","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/openai-o1-mini-dabiao\/","title":{"rendered":"OpenAI o1-mini \u5927\u6a21\u578b\u4ecb\u7ecd"},"content":{"rendered":"

\u63a8\u52a8\u9ad8\u6027\u4ef7\u6bd4\u7684\u63a8\u7406\u6280\u672f\u3002<\/p><\/blockquote>\n

\u6211\u4eec\u63a8\u51fa\u4e86 OpenAI o1-mini\uff0c\u8fd9\u662f\u4e00\u6b3e\u5177\u6709\u9ad8\u6027\u4ef7\u6bd4\u7684\u63a8\u7406\u6a21\u578b\u3002o1-mini \u5728 STEM \u9886\u57df\uff0c\u7279\u522b\u662f\u5728\u6570\u5b66\u548c\u7f16\u7a0b\u65b9\u9762\u8868\u73b0\u51fa\u8272\uff0c\u5176\u6027\u80fd\u51e0\u4e4e\u4e0e OpenAI o1<\/a>\u00a0\u5728 AIME \u548c Codeforces \u7b49\u8bc4\u6d4b\u57fa\u51c6\u4e0a\u7684\u8868\u73b0\u76f8\u5f53\u3002\u6211\u4eec\u9884\u671f\uff0co1-mini \u5c06\u6210\u4e3a\u4e00\u4e2a\u66f4\u5feb\u3001\u66f4\u7ecf\u6d4e\u7684\u9009\u62e9\uff0c\u9002\u7528\u4e8e\u90a3\u4e9b\u9700\u8981\u63a8\u7406\u4f46\u4e0d\u4f9d\u8d56\u5e7f\u6cdb\u4e16\u754c\u77e5\u8bc6\u7684\u5e94\u7528\u573a\u666f\u3002<\/p>\n

\u4eca\u5929\uff0c\u6211\u4eec\u4ee5\u6bd4 OpenAI o1-preview \u4fbf\u5b9c 80% \u7684\u4ef7\u683c\u5411 tier 5 API \u7528\u6237(opens in a new window)<\/a> \u63a8\u51fa o1-mini\u3002ChatGPT Plus\u3001Team\u3001Enterprise \u548c Edu \u7528\u6237\u53ef\u4ee5\u5c06 o1-mini \u4f5c\u4e3a o1-preview \u7684\u66ff\u4ee3\u65b9\u6848\uff0c\u4eab\u53d7\u66f4\u9ad8\u7684\u4f7f\u7528\u9650\u5236\u548c\u66f4\u4f4e\u7684\u5ef6\u8fdf\uff0c\u89c1 [Model Speed\u3002<\/p>\n

 <\/p>\n

\u4e13\u4e3a STEM \u63a8\u7406\u4f18\u5316<\/h2>\n

\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5982 o1 \u901a\u5e38\u5728\u5927\u89c4\u6a21\u6587\u672c\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u9884\u8bad\u7ec3\u3002\u5c3d\u7ba1\u8fd9\u4e9b\u9ad8\u5bb9\u91cf\u6a21\u578b\u5177\u5907\u5e7f\u6cdb\u7684\u4e16\u754c\u77e5\u8bc6\uff0c\u4f46\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u53ef\u80fd\u4f1a\u663e\u5f97\u6602\u8d35\u4e14\u901f\u5ea6\u8f83\u6162\u3002\u76f8\u5bf9\u800c\u8a00\uff0co1-mini \u662f\u4e00\u4e2a\u5728\u9884\u8bad\u7ec3\u9636\u6bb5\u4e13\u95e8\u4f18\u5316\u4e86 STEM \u63a8\u7406\u7684\u5c0f\u578b\u6a21\u578b\u3002\u5728\u4f7f\u7528\u4e0e o1 \u76f8\u540c\u7684\u9ad8\u8ba1\u7b97\u91cf\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u7ba1\u9053\u8fdb\u884c\u8bad\u7ec3\u540e\uff0co1-mini \u5728\u8bb8\u591a\u5b9e\u7528\u7684\u63a8\u7406\u4efb\u52a1\u4e0a\u8868\u73b0\u76f8\u5f53\uff0c\u540c\u65f6\u6210\u672c\u6548\u76ca\u663e\u8457\u63d0\u5347\u3002<\/p>\n

\u5728\u9700\u8981\u667a\u80fd\u548c\u63a8\u7406\u7684\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0co1-mini \u7684\u8868\u73b0\u4f18\u4e8e o1-preview \u548c o1\u3002\u7136\u800c\uff0co1-mini \u5728\u9700\u8981\u975e STEM \u4e8b\u5b9e\u77e5\u8bc6\u7684\u4efb\u52a1\u4e2d\u8868\u73b0\u8f83\u5dee\uff0c\u89c1 [Limitations]\u3002<\/p>\n

 <\/p>\n

\u6570\u5b66\u6027\u80fd\u4e0e\u63a8\u7406\u6210\u672c<\/h5>\n

\"OpenAI<\/p>\n

 <\/p>\n

\u6570\u5b66\uff1a<\/strong>\u00a0\u5728\u9ad8\u4e2d AIME \u6570\u5b66\u7ade\u8d5b\u4e2d\uff0co1-mini\uff0870.0%\uff09\u7684\u8868\u73b0\u4e0e o1\uff0874.4%\uff09\u76f8\u5f53\u2014\u2014\u800c\u4e14\u4ef7\u683c\u660e\u663e\u66f4\u4fbf\u5b9c\u2014\u2014\u5e76\u4e14\u4f18\u4e8e o1-preview\uff0844.6%\uff09\u3002o1-mini \u7684\u5f97\u5206\uff08\u5927\u7ea6\u7b54\u5bf9 11 \u9053\u9898\u76ee\u4e2d\u7684 15 \u9053\uff09\u4f7f\u5b83\u5927\u81f4\u5904\u4e8e\u7f8e\u56fd\u9ad8\u4e2d\u751f\u7684\u524d 500 \u540d\u5de6\u53f3\u3002<\/p>\n

\u7f16\u7a0b\uff1a<\/strong>\u00a0\u5728 Codeforces \u7ade\u8d5b\u7f51\u7ad9\u4e0a\uff0co1-mini \u7684 Elo \u5f97\u5206\u4e3a 1650\uff0c\u8fd9\u4e0e o1\uff081673\uff09\u76f8\u5f53\uff0c\u5e76\u4e14\u9ad8\u4e8e o1-preview\uff081258\uff09\u3002\u8fd9\u4e00 Elo \u5206\u6570\u5c06 o1-mini \u6392\u5728 Codeforces \u5e73\u53f0\u4e0a\u7a0b\u5e8f\u5458\u7684\u524d 86 \u767e\u5206\u4f4d\u3002o1-mini \u5728 HumanEval \u7f16\u7a0b\u57fa\u51c6\u6d4b\u8bd5\u548c\u9ad8\u4e2d\u6c34\u5e73\u7684\u7f51\u7edc\u5b89\u5168 Capture the Flag \u6311\u6218\uff08CTFs\uff09\u4e2d\u8868\u73b0\u4e5f\u975e\u5e38\u51fa\u8272\u3002<\/p>\n

 <\/p>\n

Codeforces<\/h5>\n
\"OpenAI<\/h5>\n

 <\/p>\n

HumanEval<\/h5>\n
\"OpenAI<\/h5>\n

 <\/p>\n

\u7f51\u7edc\u5b89\u5168 CTFs<\/h5>\n

\"OpenAI<\/strong><\/p>\n

 <\/p>\n

STEM\uff1a<\/strong>\u00a0\u5728\u4e00\u4e9b\u9700\u8981\u63a8\u7406\u7684\u5b66\u672f\u6d4b\u8bd5\u4e2d\uff0c\u4f8b\u5982 GPQA\uff08\u79d1\u5b66\uff09\u548c MATH-500\uff0co1-mini \u7684\u8868\u73b0\u8d85\u8d8a\u4e86 GPT-4o\u3002o1-mini \u5728 MMLU \u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u4e0d\u5982 GPT-4o\uff0c\u5e76\u4e14\u5728 GPQA \u4e2d\u4e5f\u843d\u540e\u4e8e o1-preview\uff0c\u56e0\u4e3a\u5b83\u7f3a\u4e4f\u5e7f\u6cdb\u7684\u4e16\u754c\u77e5\u8bc6\u3002<\/p>\n

 <\/p>\n

MMLU<\/h5>\n
\"OpenAI<\/h5>\n

 <\/p>\n

GPQA<\/h5>\n
\"OpenAI<\/h5>\n

 <\/p>\n

MATH-500<\/h5>\n

\"OpenAI<\/strong><\/p>\n

 <\/p>\n

\u4eba\u7c7b\u504f\u597d\u8bc4\u4f30\uff1a<\/strong>\u00a0\u6211\u4eec\u8ba9\u8bc4\u4f30\u8005\u5c06 o1-mini \u4e0e GPT-4o \u5728\u5404\u79cd\u9886\u57df\u7684\u5f00\u653e\u5f0f\u96be\u9898\u4e0a\u8fdb\u884c\u6bd4\u8f83\uff0c\u91c7\u7528\u4e86\u4e0e\u6211\u4eec [o1-preview \u4e0e GPT-4o \u6bd4\u8f83](https:\/\/openai.com\/index\/learning-to-reason-with-llms\/) \u76f8\u540c\u7684\u65b9\u6cd5\u3002\u7c7b\u4f3c\u4e8e o1-preview\uff0co1-mini \u5728\u9700\u8981\u63a8\u7406\u7684\u9886\u57df\u6bd4 GPT-4o \u66f4\u53d7\u6b22\u8fce\uff0c\u4f46\u5728\u8bed\u8a00\u96c6\u4e2d\u9886\u57df\u5219\u4e0d\u5982 GPT-4o \u53d7\u6b22\u8fce\u3002<\/p>\n

 <\/p>\n

\u4eba\u7c7b\u504f\u597d\u8bc4\u4f30 vs chatgpt-4o-latest<\/h5>\n

\"OpenAI<\/h2>\n

 <\/p>\n

\u6a21\u578b\u901f\u5ea6<\/h2>\n

\u4f5c\u4e3a\u5177\u4f53\u4f8b\u5b50\uff0c\u6211\u4eec\u6bd4\u8f83\u4e86 GPT-4o\u3001o1-mini \u548c o1-preview \u5728\u4e00\u4e2a\u8bcd\u6c47\u63a8\u7406\u95ee\u9898\u4e0a\u7684\u53cd\u5e94\u3002\u867d\u7136 GPT-4o \u7684\u7b54\u6848\u4e0d\u6b63\u786e\uff0c\u4f46 o1-mini \u548c o1-preview \u90fd\u7ed9\u51fa\u4e86\u6b63\u786e\u7b54\u6848\uff0c\u800c\u4e14 o1-mini \u7684\u56de\u7b54\u901f\u5ea6\u5927\u7ea6\u662f 3-5 \u500d\u5feb\u3002<\/p>\n

\"OpenAI<\/p>\n","protected":false},"excerpt":{"rendered":"

\u63a8\u52a8\u9ad8\u6027\u4ef7\u6bd4\u7684\u63a8\u7406\u6280\u672f\u3002 \u6211\u4eec\u63a8\u51fa\u4e86 OpenAI o1-mini\uff0c\u8fd9\u662f\u4e00\u6b3e\u5177\u6709\u9ad8\u6027\u4ef7\u6bd4\u7684\u63a8\u7406\u6a21\u578b\u3002o1-mini \u5728 STEM \u9886\u57df\uff0c\u7279\u522b\u662f\u5728\u6570\u5b66\u548c\u7f16\u7a0b\u65b9\u9762\u8868\u73b0\u51fa\u8272\uff0c\u5176\u6027\u80fd\u51e0\u4e4e\u4e0e OpenAI o1\u00a0\u5728 AIME \u548c Codeforces…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[34],"tags":[],"class_list":["post-6029","post","type-post","status-publish","format-standard","hentry","category-knowledge"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/6029","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/comments?post=6029"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/6029\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=6029"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=6029"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=6029"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}