{"id":1210,"date":"2025-07-21T17:08:00","date_gmt":"2025-07-21T09:08:00","guid":{"rendered":"http:\/\/www.applicationsofllm.com\/?p=1210"},"modified":"2025-09-17T17:14:33","modified_gmt":"2025-09-17T09:14:33","slug":"deepseek%e3%80%81gpt-5%e5%b8%a6%e5%a4%b4%e8%bd%ac%e5%90%91%e6%b7%b7%e5%90%88%e6%8e%a8%e7%90%86%ef%bc%8c%e4%b8%80%e4%b8%aatoken%e4%b9%9f%e4%b8%8d%e8%83%bd%e6%b5%aa%e8%b4%b9","status":"publish","type":"post","link":"http:\/\/www.applicationsofllm.com\/index.php\/2025\/07\/21\/deepseek%e3%80%81gpt-5%e5%b8%a6%e5%a4%b4%e8%bd%ac%e5%90%91%e6%b7%b7%e5%90%88%e6%8e%a8%e7%90%86%ef%bc%8c%e4%b8%80%e4%b8%aatoken%e4%b9%9f%e4%b8%8d%e8%83%bd%e6%b5%aa%e8%b4%b9\/","title":{"rendered":"DeepSeek\u3001GPT-5\u5e26\u5934\u8f6c\u5411\u6df7\u5408\u63a8\u7406\uff0c\u4e00\u4e2atoken\u4e5f\u4e0d\u80fd\u6d6a\u8d39"},"content":{"rendered":"\n

\u5728\u6700\u8fd1\u7684\u4e00\u6863\u8131\u53e3\u79c0\u8282\u76ee\u4e2d\uff0c\u6f14\u5458\u5f20\u4fca\u8c03\u4f83 DeepSeek \u662f\u4e00\u6b3e\u975e\u5e38\u300c\u5185\u8017\u300d\u7684 AI\uff0c\u8fde\u4e2a\u300c1 \u52a0 1 \u7b49\u4e8e\u51e0\u300d\u90fd\u8981\u659f\u914c\u534a\u5929\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5728 AI \u9886\u57df\uff0c\u8fd9\u79cd\u60c5\u51b5\u88ab\u79f0\u4e3a\u300c\u8fc7\u5ea6\u601d\u8003<\/strong>\u300d\u3002\u5b83\u7684\u5b58\u5728\u8ba9 AI \u5927\u6a21\u578b\u516c\u53f8\u975e\u5e38\u5934\u75bc\uff0c\u56e0\u4e3a\u5b9e\u5728\u662f\u592a\u6d6a\u8d39\u7b97\u529b\u4e86\uff0c\u90a3\u70b9\u8ba2\u9605\u8d39\u6839\u672c cover \u4e0d\u4f4f\u3002<\/p>\n\n\n\n

\u6240\u4ee5\uff0c\u65e9\u5728\u53bb\u5e74\u7684 GTC \u5927\u4f1a\u4e0a\uff0cTransformer \u8bba\u6587\u4f5c\u8005\u4e4b\u4e00 Illia Polosukhin \u5c31\u63d0\u5230\uff0c\u81ea\u9002\u5e94\u8ba1\u7b97<\/strong>\u662f\u63a5\u4e0b\u6765\u5fc5\u987b\u51fa\u73b0\u7684\u4e8b\u60c5\u4e4b\u4e00\uff0c\u6211\u4eec\u9700\u8981\u77e5\u9053\u5728\u7279\u5b9a\u95ee\u9898\u4e0a\u5e94\u8be5\u82b1\u8d39\u591a\u5c11\u8ba1\u7b97\u8d44\u6e90\u3002<\/p>\n\n\n\n

\u4eca\u5e74\uff0c\u8d8a\u6765\u8d8a\u591a\u7684\u6a21\u578b\u5382\u5546\u5c06\u8fd9\u4ef6\u4e8b\u63d0\u4e0a\u65e5\u7a0b \u2014\u2014OpenAI \u7ed9 GPT-5 \u88c5\u4e86\u4e2a\u300c\u8def\u7531\u5668\u300d\uff0c\u786e\u4fdd\u6a21\u578b\u53ef\u4ee5\u5728\u62ff\u5230\u7528\u6237\u95ee\u9898\u540e\uff0c\u81ea\u52a8\u9009\u62e9\u5408\u9002\u7684\u6a21\u578b\uff0c\u50cf\u300c\u5929\u7a7a\u4e3a\u4ec0\u4e48\u662f\u84dd\u8272\u7684\u300d\u8fd9\u79cd\u95ee\u9898\u76f4\u63a5\u5c31\u4e22\u7ed9\u8f7b\u91cf\u7ea7\u6a21\u578b\uff1bDeepSeek \u66f4\u6fc0\u8fdb\uff0c\u76f4\u63a5\u628a\u5bf9\u8bdd\u548c\u63a8\u7406\u80fd\u529b\u5408\u5e76\u5230\u4e86\u4e00\u4e2a\u6a21\u578b\u91cc\uff0c\u63a8\u51fa\u4e86\u5355\u6a21\u578b\u53cc\u6a21\u5f0f\u7684 DeepSeek v3.1\u3002<\/p>\n\n\n\n

\u5982\u56fe\u6240\u793a\uff0c\u8fd9\u4e24\u79cd\u65b9\u6848\u5728\u8282\u7701 token \u65b9\u9762\u90fd\u6709\u663e\u8457\u7684\u6548\u679c\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5728\u5185\u90e8\u8bc4\u6d4b\u4e2d\uff0cGPT-5\uff08\u4f7f\u7528\u601d\u8003\u6a21\u5f0f\uff09\u80fd\u4ee5\u6bd4\u524d\u4ee3\u6a21\u578b\u66f4\u5c11\u7684 token \u6570\u5b8c\u6210\u4efb\u52a1 \u2014\u2014 \u5927\u7ea6\u5c11 50\u201380% \u7684\u8f93\u51fa token \u5373\u53ef\u8fbe\u5230\u76f8\u540c\u751a\u81f3\u66f4\u597d\u7684\u6548\u679c\u3002<\/sup><\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u6d4b\u8bd5\u6570\u636e\u663e\u793a\uff0c\u5728 AIME 2025\u3001GPQA Diamond \u548c LiveCodeBench \u8fd9\u4e9b\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cDeepSeek v3.1\uff08\u4f7f\u7528\u601d\u8003\u6a21\u5f0f\uff09\u548c DeepSeek R1 \u5f97\u5206\u7c7b\u4f3c\uff0c\u4f46\u6d88\u8017\u7684 token \u6570\u91cf\u51cf\u5c11\u4e86 25-50%\u3002<\/sup><\/p>\n\n\n\n

\u672a\u6765\u4e00\u6bb5\u65f6\u95f4\uff0c\u8fd9\u79cd\u6df7\u5408\u63a8\u7406\u6a21\u5f0f<\/strong>\u6709\u671b\u6210\u4e3a\u5927\u6a21\u578b\u9886\u57df\u7684\u65b0\u5e38\u6001\u3002\u5982\u4f55\u5728\u6210\u672c\u548c\u6027\u80fd\u4e4b\u95f4\u53d6\u5f97\u5e73\u8861\u6b63\u6210\u4e3a\u6a21\u578b\u7ade\u4e89\u529b\u7684\u65b0\u57fa\u51c6\u3002<\/p>\n\n\n\n

\u5728\u8fd9\u7bc7\u6587\u7ae0\u4e2d\uff0c\u6211\u4eec\u5c06\u8ba8\u8bba\u8fd9\u79cd\u8d8b\u52bf\u7684\u6210\u56e0\u3001\u5404\u5927\u516c\u53f8\u7684\u52a8\u5411\u4ee5\u53ca\u76f8\u5173\u7684\u7814\u7a76\uff0c\u5e0c\u671b\u5bf9\u5927\u5bb6\u6709\u6240\u542f\u53d1\u3002<\/p>\n\n\n\n

\u6700\u597d\u7684\u6a21\u578b\u6c38\u8fdc\u6700\u53d7\u6b22\u8fce  \u4f46\u6a21\u578b\u5382\u5546\u600e\u4e48 cover \u6210\u672c\uff1f <\/strong> <\/p>\n\n\n\n

\u524d\u6bb5\u65f6\u95f4\uff0cTextQL \u8054\u5408\u521b\u59cb\u4eba\u517c CEO \u4e01\u4e00\u5e06\uff08Ethan Ding\uff09\u5728\u4e00\u7bc7\u4e2d\u6df1\u5165\u8ba8\u8bba\u4e86\u4e00\u4e2a\u53cd\u76f4\u89c9\u7684\u73b0\u8c61 \u2014\u2014 \u660e\u660e Token \u7684\u6210\u672c\u4e00\u76f4\u5728\u4e0b\u964d\uff0c\u4f46\u5404\u5bb6\u6a21\u578b\u516c\u53f8\u7684\u8ba2\u9605\u8d39\u5374\u5728\u98de\u6da8\u3002<\/p>\n\n\n\n

\u4ed6\u5c06\u8fd9\u4e00\u95ee\u9898\u7684\u75c7\u7ed3\u5f52\u7ed3\u4e3a\uff1a\u90a3\u4e9b\u964d\u4ef7\u7684\u6a21\u578b\u5927\u90e8\u5206\u4e0d\u662f SOTA \u6a21\u578b\uff0c\u800c\u4eba\u7c7b\u5728\u8ba4\u77e5\u4e0a\u7684\u8d2a\u5a6a\u51b3\u5b9a\u4e86\uff0c\u5927\u90e8\u5206\u4eba\u53ea\u60f3\u8981\u300c\u6700\u5f3a\u5927\u8111\u300d\uff0c\u6240\u4ee5 99% \u7684\u9700\u6c42\u4f1a\u8f6c\u5411 SOTA\u3002\u800c\u6700\u5f3a\u6a21\u578b\u7684\u4ef7\u683c\u59cb\u7ec8\u5dee\u4e0d\u591a\u3002<\/p>\n\n\n\n

\u66f4\u7cdf\u7cd5\u7684\u662f\uff0c\u968f\u7740\u300c\u6df1\u5ea6\u7814\u7a76\u300d\u3001Agent \u7b49\u6a21\u5f0f\u7684\u51fa\u73b0\uff0cAI \u80fd\u5b8c\u6210\u7684\u4efb\u52a1\u957f\u5ea6\u6bcf 6 \u4e2a\u6708\u5c31\u7ffb\u4e00\u500d\u3002\u5230 2027 \u5e74\uff0c\u6211\u4eec\u53ef\u80fd\u5c06\u62e5\u6709\u80fd\u8fde\u7eed\u8fd0\u884c 24 \u5c0f\u65f6\u3001\u800c\u4e14\u4e0d\u4f1a\u8dd1\u9898\u7684 AI agent\u3002\u6309\u7167\u8fd9\u4e00\u8d8b\u52bf\u53d1\u5c55\u4e0b\u53bb\uff0c\u8fd9\u4e9b\u300c\u6700\u5f3a\u5927\u8111\u300d\u6240\u6d88\u8017\u7684 token \u6570\u91cf\u5c06\u4f1a\u7206\u70b8\u5f0f\u589e\u957f\u3002<\/p>\n\n\n\n

\u6362\u7b97\u6210\u7ecf\u6d4e\u8d26\uff0c\u8fd9\u610f\u5473\u7740\uff0c\u73b0\u5728\u4e00\u6b21 20 \u5206\u949f\u7684\u300c\u6df1\u5ea6\u7814\u7a76\u300d\u8c03\u7528\u5927\u6982\u82b1\u8d39 1 \u7f8e\u5143\uff0c\u4f46\u5230\u4e86 2027 \u5e74\uff0c\u4e00\u6b21 Agent \u8c03\u7528\u5c31\u53d8\u6210\u4e86 72 \u7f8e\u5143 \/ \u5929 \/ \u7528\u6237\u3002<\/p>\n\n\n\n

\u6240\u4ee5\uff0c\u4eca\u5e74\u597d\u591a AI \u6a21\u578b\u5382\u5546\u90fd\u63d0\u9ad8\u4e86\u8ba2\u9605\u8d39\uff0c\u8fd8\u9650\u5236\u7528\u91cf\u3002\u56e0\u4e3a\u539f\u6765\u6bcf\u6708 20 \u7f8e\u5143\u7684\u8ba2\u9605\u8d39\uff0c\u8fde\u7528\u6237\u6bcf\u5929\u8fdb\u884c\u4e00\u6b21 1 \u7f8e\u5143\u7684\u6df1\u5ea6\u8c03\u7528\u90fd\u6491\u4e0d\u8d77\u3002<\/p>\n\n\n\n

\u8fd9\u90e8\u5206\u591a\u51fa\u6765\u7684\u8ba2\u9605\u8d39\u7ed9\u6a21\u578b\u5382\u5546\u63d0\u4f9b\u4e86\u4e00\u4e9b\u7f13\u51b2\u7a7a\u95f4\uff0c\u4f46\u7ec8\u7a76\u662f\u7f13\u5175\u4e4b\u8ba1\u3002\u6240\u4ee5\u6a21\u578b\u5382\u5546\u8fd8\u60f3\u4e86\u4e00\u4e9b\u5176\u4ed6\u7684\u529e\u6cd5\u6765\u5e94\u5bf9\u6210\u672c\u79ef\u538b\uff0c\u5305\u62ec\u5c06\u5904\u7406\u4efb\u52a1\u5378\u8f7d\u5230\u7528\u6237\u673a\u5668\u4e0a\u3001\u6839\u636e\u8d1f\u8f7d\u81ea\u52a8\u5207\u6362\u6a21\u578b\u7b49\u3002\u6211\u4eec\u5728 GPT-5 \u4e2d\u770b\u5230\u5c31\u662f\u540e\u9762\u8fd9\u79cd\u505a\u6cd5\u3002DeepSeek \u5219\u66f4\u8fdb\u4e00\u6b65\uff0c\u8ba9\u4e00\u4e2a\u6a21\u578b\u5224\u65ad\u95ee\u9898\u96be\u5ea6\uff0c\u7136\u540e\u5728\u601d\u8003 \/ \u975e\u601d\u8003\u6a21\u5f0f\u4e4b\u95f4\u5207\u6362\u3002\u9664\u6b64\u4e4b\u5916\uff0cClaude\u3001Qwen \u7b49\u4e5f\u662f\u8fd9\u6761\u8def\u7ebf\u7684\u63a2\u7d22\u8005\uff0c\u540c\u6837\u503c\u5f97\u5173\u6ce8\u3002<\/p>\n\n\n\n

\u8fd9\u4e9b\u5927\u6a21\u578b\u90fd\u5728\u5c1d\u8bd5\u6df7\u5408\u63a8\u7406<\/strong><\/p>\n\n\n\n

Anthropic \u7684 Claude \u7cfb\u5217<\/strong><\/p>\n\n\n\n

Anthropic \u4eca\u5e74 2 \u6708\u4efd\u63a8\u51fa\u7684 Claude 3.7 Sonnet <\/strong>\u662f\u5e02\u573a\u4e0a\u9996\u4e2a\u6df7\u5408\u63a8\u7406\u6a21\u578b\u3002\u5b83\u53ef\u4ee5\u8fd1\u4e4e\u5b9e\u65f6\u5730\u7ed9\u51fa\u56de\u5e94\uff0c\u4e5f\u53ef\u4ee5\u8fdb\u884c\u6df1\u5165\u7684\u3001\u9010\u6b65\u5c55\u5f00\u7684\u601d\u8003\uff0c\u5e76\u5c06\u601d\u8003\u8fc7\u7a0b\u5c55\u793a\u7ed9\u7528\u6237\u3002API \u7528\u6237\u8fd8\u80fd\u7cbe\u7ec6\u63a7\u5236\u6a21\u578b\u7684\u601d\u8003\u65f6\u957f\uff08\u8ba9 Claude \u601d\u8003\u4e0d\u8d85\u8fc7 N \u4e2a token\uff09\u3002<\/p>\n\n\n\n

\u5728\u5f53\u65f6\u7684\u535a\u5ba2\u91cc\uff0cAnthropic \u5c31\u89e3\u91ca\u4e86\u4ed6\u4eec\u7684\u7406\u5ff5\uff1a\u300c\u6211\u4eec\u5f00\u53d1 Claude 3.7 Sonnet \u7684\u7406\u5ff5\u4e0e\u5e02\u9762\u4e0a\u5176\u4ed6\u63a8\u7406\u6a21\u578b\u622a\u7136\u4e0d\u540c\u3002\u6b63\u5982\u4eba\u7c7b\u4f7f\u7528\u5355\u4e2a\u5927\u8111\u8fdb\u884c\u5feb\u901f\u54cd\u5e94\u548c\u6df1\u5ea6\u601d\u8003\u4e00\u6837\uff0c\u6211\u4eec\u8ba4\u4e3a\u63a8\u7406\u5e94\u8be5\u662f\u524d\u6cbf\u6a21\u578b\u7684\u96c6\u6210\u80fd\u529b\uff0c\u800c\u975e\u4e00\u4e2a\u5b8c\u5168\u72ec\u7acb\u7684\u6a21\u578b<\/strong>\u3002\u8fd9\u79cd\u7edf\u4e00\u7684\u65b9\u6cd5\u4e5f\u4e3a\u7528\u6237\u5e26\u6765\u4e86\u66f4\u6d41\u7545\u7684\u4f53\u9a8c\u3002\u300d<\/p>\n\n\n\n

\u5728\u4e4b\u540e\u7684 Claude 4 \u7cfb\u5217\u6a21\u578b\u4e2d\uff0cAnthropic \u5ef6\u7eed\u4e86\u8fd9\u79cd\u6a21\u5f0f\u3002\u4e0d\u8fc7\uff0c\u4ed6\u4eec\u4e00\u76f4\u4fdd\u7559\u4e86\u4e00\u4e2a\u300c\u6269\u5c55\u601d\u8003\u300d\u7684\u5f00\u5173\uff0c\u8ba9\u7528\u6237\u6765\u51b3\u5b9a\u4f55\u65f6\u5f00\u542f\u6df1\u5ea6\u601d\u8003\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u963f\u91cc\u7684 Qwen3 \u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u963f\u91cc\u4eca\u5e74 4 \u6708\u4efd\u5f00\u6e90\u7684 Qwen3<\/strong> \u7cfb\u5217\u6a21\u578b\u662f\u6df7\u5408\u63a8\u7406\u6a21\u578b\u7684\u5f00\u6e90\u4ee3\u8868\uff0c\u91c7\u7528\u5355\u4e00\u6a21\u578b\u6846\u67b6\u878d\u5408\u4e86\u601d\u8003\u6a21\u5f0f\u4e0e\u975e\u601d\u8003\u6a21\u5f0f\u3002\u4e24\u79cd\u6a21\u5f0f\u7684\u5207\u6362\u5b8c\u5168\u7531\u7528\u6237\u63a7\u5236\uff0c\u4e0d\u4f9d\u8d56\u4e8e\u81ea\u52a8\u68c0\u6d4b\u6216\u5176\u4ed6\u9690\u5f0f\u89e6\u53d1\u3002<\/p>\n\n\n\n

\u5177\u4f53\u6765\u8bf4\uff0c\u5b83\u652f\u6301\u5728\u5bf9\u8bdd\u4e2d\u63d2\u5165\u7279\u6b8a\u6807\u8bb0 \/think \u6216 \/no_think \u6765\u52a8\u6001\u5207\u6362\uff0c\u6216\u8005\u5728 API \u8c03\u7528\u65f6\u8bbe\u7f6e\u7279\u5b9a\u53c2\u6570\u3002<\/p>\n\n\n\n

\u4e3a\u9632\u6b62\u63a8\u7406\u8fc7\u7a0b\u8fc7\u957f\uff0cQwen 3 \u8fd8\u63d0\u4f9b\u4e86 thinking_budget \u53c2\u6570\uff0c\u7528\u6237\u53ef\u4ee5\u8bbe\u5b9a\u63a8\u7406\u94fe\u6700\u5927\u7684 token \u6570\uff1b\u82e5\u5b9e\u9645\u63a8\u7406\u8d85\u8fc7\u6b64\u9884\u7b97\uff0c\u6a21\u578b\u4f1a\u622a\u65ad\u4e2d\u95f4\u5185\u5bb9\u5e76\u76f4\u63a5\u751f\u6210\u6700\u7ec8\u7b54\u6848\u3002<\/p>\n\n\n\n

\u5177\u4f53\u6280\u672f\u4fe1\u606f\u53ef\u4ee5\u53c2\u89c1 Qwen 3 \u6280\u672f\u62a5\u544a\uff1ahttps:\/\/arxiv.org\/pdf\/2505.09388<\/p>\n\n\n\n

\u4e0d\u8fc7\uff0c\u8fd9\u4e2a\u7cfb\u5217\u7684\u6df7\u5408\u63a8\u7406\u6a21\u578b\u5e76\u6ca1\u6709\u8fbe\u5230\u7406\u60f3\u6548\u679c\uff0c\u5728\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4e5f\u4e0d\u591f\u597d\u3002\u6240\u4ee5\u5728\u4e0e\u793e\u533a\u6c9f\u901a\u5e76\u6df1\u601d\u719f\u8651\u540e\uff0c\u963f\u91cc\u51b3\u5b9a\u505c\u7528\u8be5\u6a21\u5f0f\uff0c\u8f6c\u5934\u5206\u522b\u8bad\u7ec3 Instruct \u6a21\u578b\u548c Thinking \u6a21\u578b\u3002\u65b0\u6a21\u578b\u5df2\u7ecf\u5728 7 \u6708\u4efd\u6b63\u5f0f\u5f00\u6e90\uff0c\u5e76\u4e14\u76f8\u6bd4\u6df7\u5408\u63a8\u7406\u6a21\u578b\u5b9e\u73b0\u4e86\u660e\u663e\u7684\u6027\u80fd\u63d0\u5347\uff08\u5c24\u5176\u662f instruct \u6a21\u578b\uff09\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5bf9\u4e8e Qwen \u6765\u8bf4\uff0c\u8fd9\u7b97\u662f\u4e00\u4e2a\u5c0f\u5c0f\u7684\u632b\u6298\u3002\u4f46\u8be5\u56e2\u961f\u5e76\u6ca1\u6709\u5b8c\u5168\u653e\u5f03\u8fd9\u4e2a\u60f3\u6cd5\u3002\u300c\u6211\u4eec\u4ecd\u5728\u7ee7\u7eed\u7814\u7a76\u6df7\u5408\u601d\u7ef4\u6a21\u5f0f\uff0c\u300d\u8be5\u56e2\u961f\u5199\u9053\uff0c\u5e76\u6697\u793a\u4e00\u65e6\u89e3\u51b3\u4e86\u8d28\u91cf\u95ee\u9898\uff0c\u8be5\u529f\u80fd\u53ef\u80fd\u4f1a\u5728\u672a\u6765\u7684\u6a21\u578b\u4e2d\u91cd\u65b0\u51fa\u73b0\u3002<\/p>\n\n\n\n

\u8c37\u6b4c\u7684 Gemini \u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u4eca\u5e74 4 \u6708\uff0c\u8c37\u6b4c\u63a8\u51fa\u4e86\u9996\u6b3e\u6df7\u5408\u63a8\u7406\u6a21\u578b \u2014\u2014Gemini 2.5 Flash<\/strong>\u3002\u8be5\u6a21\u578b\u5f15\u5165\u4e86\u300c\u601d\u8003\u9884\u7b97<\/strong>\u300d\u673a\u5236\uff0c\u5141\u8bb8\u5f00\u53d1\u4eba\u5458\u6307\u5b9a\u5728\u751f\u6210\u54cd\u5e94\u4e4b\u524d\u5e94\u5206\u914d\u591a\u5c11\u8ba1\u7b97\u80fd\u529b\u7528\u4e8e\u63a8\u7406\u590d\u6742\u95ee\u9898\u3002\u6a21\u578b\u5728\u751f\u6210\u54cd\u5e94\u4e4b\u524d\u4f1a\u8bc4\u4f30\u591a\u79cd\u6f5c\u5728\u8def\u5f84\u548c\u8003\u8651\u56e0\u7d20\u3002\u601d\u8003\u9884\u7b97\u53ef\u4ee5\u4ece 0 \u8c03\u6574\u5230 24,576 \u4e2a token\u3002\u4f7f\u7528 Gemini 2.5 Flash \u65f6\uff0c\u8f93\u51fa\u6210\u672c\u4f1a\u56e0\u662f\u5426\u542f\u7528\u63a8\u7406\u529f\u80fd\u76f8\u5dee 6 \u500d\u3002<\/p>\n\n\n\n

\u66f4\u64c5\u957f\u6df1\u5ea6\u601d\u8003\u7684 Gemini 2.5 Pro \u867d\u7136\u5728\u53d1\u5e03\u65f6\u6ca1\u6709\u300c\u601d\u8003\u9884\u7b97\u300d\u673a\u5236\uff0c\u4f46\u5728 6 \u6708\u4efd\u7684\u4e00\u6b21\u91cd\u5927\u66f4\u65b0\u65f6\u53c8\u52a0\u4e0a\u4e86\u3002<\/p>\n\n\n\n

\u5b83\u7684\u51fa\u73b0\u5219\u88ab\u5b9a\u4e49\u4e3a\u9762\u5411 B \u7aef\u7684\u5b9e\u7528\u4e3b\u4e49\u521b\u65b0\uff0c\u800c\u975e\u4e00\u4e2a\u9762\u5411\u666e\u901a\u6d88\u8d39\u8005\u7684\u901a\u7528\u6a21\u578b\u3002\u56e0\u4e3a\u5b83\u5141\u8bb8\u4f01\u4e1a\u5728\u751f\u4ea7\u7cfb\u7edf\u4e2d\u50cf\u8c03\u8282\u6c34\u9f99\u5934\u4e00\u6837\u7cbe\u786e\u8c03\u8282 AI \u7684\u601d\u8003\u6210\u672c\uff0c\u8fd9\u5bf9\u4e8e\u9700\u8981\u5c06 AI \u5e94\u7528\u5927\u89c4\u6a21\u90e8\u7f72\u7684\u4f01\u4e1a\u548c\u5f00\u53d1\u8005\u6765\u8bf4\u662f\u4e00\u4e2a\u975e\u5e38\u4f1f\u5927\u7684\u529f\u80fd\u3002<\/p>\n\n\n\n

\u5728\u5b9e\u73b0\u65b9\u5f0f\u4e0a\uff0c\u6709\u4eba\u731c\u6d4b\u8fd9\u53ef\u80fd\u662f\u4e00\u4e2a\u300c\u6df7\u5408\u65b9\u6848\u300d\u2014\u2014 \u6a21\u578b\u53ef\u80fd\u5b9e\u9645\u7ed3\u5408\u4e86\u4e00\u4e2a\u64c5\u957f\u63a8\u7406\u7684\u5927\u6a21\u578b\u548c\u4e00\u4e2a\u7528\u4e8e\u8f93\u51fa\u7684\u5c0f\u6a21\u578b\uff0c\u4e24\u8005\u6839\u636e\u9884\u7b97\u5207\u6362\u3002\u4e0d\u8fc7\uff0c\u8fd9\u4e2a\u731c\u60f3\u8fd8\u672a\u88ab\u8bc1\u5b9e\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Gemini 2.5 \u7cfb\u5217\u6280\u672f\u62a5\u544a\uff1ahttps:\/\/arxiv.org\/pdf\/2507.06261<\/p>\n\n\n\n

\u5feb\u624b\u7684 Kwai \u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u5feb\u624b\u4e8e\u4eca\u5e74 6 \u6708\u521d\u63a8\u51fa\u4e86\u81ea\u52a8\u601d\u8003\u5927\u6a21\u578b KwaiCoder-AutoThink-preview<\/strong>\u3002\u8be5\u6a21\u578b\u878d\u5408\u4e86\u300c\u601d\u8003\u300d\u548c\u300c\u975e\u601d\u8003\u300d\u80fd\u529b\uff0c\u5e76\u5177\u5907\u6839\u636e\u95ee\u9898\u96be\u5ea6\u81ea\u52a8\u5207\u6362\u601d\u8003\u5f62\u6001\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n

\u4ed6\u4eec\u7684\u6838\u5fc3\u60f3\u6cd5\u662f\u5728\u601d\u8003\u4e4b\u524d\u52a0\u4e0a\u4e00\u4e2a pre-think \u7684\u9636\u6bb5\uff0c\u8ba9\u6a21\u578b\u9884\u5148\u5224\u65ad\u95ee\u9898\u7684\u56f0\u96be\u5ea6\u3002<\/p>\n\n\n\n

\u7b80\u5355\u6765\u8bf4\uff0cKwaiCoder-AutoThink-preview \u6a21\u578b\u91c7\u7528\u4e86\u4e24\u6b65\u5f0f\u8bad\u7ec3\u65b9\u6cd5\uff0c\u9996\u5148\u901a\u8fc7 Agentic \u65b9\u6cd5\u6784\u9020\u957f\u77ed\u601d\u8003\u7684 Cold Start \u6570\u636e\u8ba9\u6a21\u578b\u5728\u8fdb\u884c\u601d\u8003\u4e4b\u524d\u5148\u8fdb\u884c\u4e00\u4e2a\u300cpre-think\u300d\uff0c\u5224\u65ad\u4e00\u4e0b\u95ee\u9898\u7684\u96be\u5ea6\u3002 \u7136\u540e\u518d\u4f7f\u7528\u52a0\u4e0a\u4e13\u95e8\u4e3a Auto Think \u4efb\u52a1\u8bbe\u8ba1\u7684\u5e26\u6709\u8fc7\u7a0b\u76d1\u7763\u7684 Step-SRPO \u589e\u5f3a\u6a21\u578b\u5bf9\u5404\u79cd\u4efb\u52a1\u96be\u4ee5\u7a0b\u5ea6\u5224\u65ad\u7684\u51c6\u786e\u6027\u3002<\/p>\n\n\n\n

\u4eca\u5e74 7 \u6708\uff0c\u5feb\u624b\u66f4\u8fdb\u4e00\u6b65\uff0c\u5f00\u6e90\u4e86 AutoThink \u5927\u6a21\u578b KAT-V1<\/strong>\uff0c\u4e5f\u662f\u4e3b\u6253\u65e0\u9700\u4eba\u7c7b\u624b\u52a8\u8bbe\u7f6e\uff0c\u6a21\u578b\u81ea\u4e3b\u5224\u65ad\u3002<\/p>\n\n\n\n

\u5177\u4f53\u7ec6\u8282\u53ef\u4ee5\u53c2\u89c1\u6280\u672f\u62a5\u544a\u3002<\/p>\n\n\n\n

\u6280\u672f\u62a5\u544a\uff1ahttps:\/\/arxiv.org\/pdf\/2507.08297<\/p>\n\n\n\n

\u5b57\u8282\u7684\u8c46\u5305\u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u5b57\u8282\u4eca\u5e74 6 \u6708\u53d1\u5e03\u7684 Seed 1.6 (Adaptive CoT) <\/strong>\u4e5f\u662f\u4e00\u4e2a\u6df7\u5408\u63a8\u7406\u6a21\u578b\uff0c\u652f\u6301 on\/off\/auto \u4e09\u79cd\u601d\u8003\u6a21\u5f0f\uff0c\u8ba9\u7528\u6237\u53ef\u4ee5\u6839\u636e\u4f7f\u7528\u573a\u666f\u81ea\u884c\u9009\u62e9\uff0c\u5927\u6a21\u578b\u4e5f\u53ef\u4ee5\u81ea\u5df1\u5224\u65ad\u662f\u5426\u4f7f\u7528\u6df1\u5ea6\u601d\u8003\u3002<\/p>\n\n\n\n

\u636e\u5b98\u65b9\u4ecb\u7ecd\uff0c\u8fd9\u79cd\u81ea\u9002\u5e94\u601d\u8003\u80fd\u529b\u7684\u5b9e\u73b0\u4f9d\u9760\u4e00\u79cd\u52a8\u6001\u601d\u8003\u6280\u672f\u6765\u5b9e\u73b0\uff0c\u5373 Adaptive CoT\uff0c\u80fd\u5728\u4fdd\u8bc1\u6548\u679c\u7684\u540c\u65f6\u538b\u7f29 CoT \u957f\u5ea6\u3002<\/p>\n\n\n\n

Adaptive CoT \u76f8\u5173\u8bba\u6587\u5728 5 \u6708\u4efd\u5c31\u5df2\u7ecf\u4e0a\u7ebf\uff08AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learning\uff09\uff0c\u5b83\u5c06\u81ea\u9002\u5e94\u63a8\u7406\u5efa\u6a21\u4e3a\u4e00\u4e2a\u5e15\u7d2f\u6258\u4f18\u5316\u95ee\u9898\uff1a\u5728\u4fdd\u8bc1\u6a21\u578b\u6027\u80fd\u7684\u540c\u65f6\uff0c\u6700\u5c0f\u5316 CoT \u8c03\u7528\u5e26\u6765\u7684\u6210\u672c\uff08\u5305\u62ec\u89e6\u53d1\u9891\u6b21\u4e0e\u8ba1\u7b97\u5f00\u9500\uff09\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u7814\u7a76\u8005\u91c7\u7528\u57fa\u4e8e\u5f3a\u5316\u5b66\u4e60\u7684\u65b9\u6cd5\uff0c\u4f7f\u7528\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\uff08PPO\uff09\uff0c\u901a\u8fc7\u52a8\u6001\u8c03\u6574\u60e9\u7f5a\u7cfb\u6570\u6765\u63a7\u5236 CoT \u89e6\u53d1\u51b3\u7b56\u8fb9\u754c\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u4f9d\u636e\u9690\u542b\u7684\u95ee\u9898\u590d\u6742\u5ea6\u5224\u65ad\u662f\u5426\u9700\u8981 CoT\u3002\u5173\u952e\u6280\u672f\u8d21\u732e\u4e4b\u4e00\u662f\u300c\u9009\u62e9\u6027\u635f\u5931\u63a9\u853d\u300d\uff08Selective Loss Masking\uff0cSLM\uff09\uff0c\u7528\u4ee5\u9632\u6b62\u591a\u9636\u6bb5 RL \u8bad\u7ec3\u4e2d\u7684\u51b3\u7b56\u8fb9\u754c\u5d29\u584c\uff0c\u786e\u4fdd\u89e6\u53d1\u673a\u5236\u7a33\u5065\u4e14\u7a33\u5b9a\u3002\u5f53\u65f6\uff0c\u8fd9\u9879\u6280\u672f\u9996\u5148\u88ab\u90e8\u7f72\u5230\u4e86 doubao-1.5-thinking-pro-m-250428 \u7248\u672c\u91cc\u3002<\/p>\n\n\n\n

\u5177\u4f53\u7ec6\u8282\u53ef\u53c2\u89c1\u8bba\u6587\uff1ahttps:\/\/arxiv.org\/pdf\/2505.11896<\/p>\n\n\n\n

\u4e0d\u8fc7\uff0c\u5b57\u8282\u8868\u793a\uff0c\u4ed6\u4eec\u6700\u7ec8\u8fd8\u662f\u5e0c\u671b\u5c06\uff08Seed1.6-Thinking \u6240\u4ee3\u8868\u7684\uff09\u6781\u81f4\u63a8\u7406\u6548\u679c\u548c\uff08Seed 1.6 \u6240\u4ee3\u8868\u7684\uff09\u52a8\u6001\u601d\u8003\u6280\u672f\u878d\u5408\u5230\u4e00\u4e2a\u6a21\u578b\u91cc\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u66f4\u667a\u80fd\u7684\u6a21\u578b\u3002<\/p>\n\n\n\n

\u817e\u8baf\u7684\u6df7\u5143\u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u817e\u8baf\u4eca\u5e74 6 \u6708\u4efd\u53d1\u5e03\u7684 Hunyuan-A13B<\/strong> \u4e5f\u662f\u4e00\u4e2a\u6df7\u5408\u63a8\u7406\u6a21\u578b\u3002\u4e3a\u4e86\u8ba9\u6a21\u578b\u57fa\u4e8e\u4efb\u52a1\u9700\u6c42\u52a8\u6001\u8c03\u6574\u63a8\u7406\u6df1\u5ea6\uff0c\u4ed6\u4eec\u5b9e\u73b0\u4e86\u4e00\u4e2a\u53cc\u6a21\u5f0f\u601d\u7ef4\u94fe\uff08Dual-Mode CoT\uff09\u6846\u67b6\uff0c\u8ba9\u6a21\u578b\u5728\u5feb\u3001\u6162\u601d\u8003\u4e4b\u95f4\u5207\u6362\u3002<\/p>\n\n\n\n

\u5728\u6280\u672f\u62a5\u544a\u4e2d\uff0c\u4ed6\u4eec\u63d0\u5230\u4e86\u8fd9\u4e2a\u6846\u67b6\u7684\u4e00\u4e9b\u7ec6\u8282\u3002\u5728\u540e\u8bad\u7ec3\u9636\u6bb5\uff0c\u4ed6\u4eec\u91c7\u7528\u7edf\u4e00\u7684\u8bad\u7ec3\u7ed3\u6784\u6765\u540c\u65f6\u4f18\u5316\u4e24\u79cd\u63a8\u7406\u6a21\u5f0f\u3002\u4e3a\u4e86\u4f7f\u6a21\u578b\u8f93\u51fa\u6807\u51c6\u5316\uff0c\u4e24\u79cd\u6a21\u5f0f\u7684\u8bad\u7ec3\u6837\u672c\u5747\u91c7\u7528\u7edf\u4e00\u7ed3\u6784\u5316\u8bbe\u8ba1\uff1a\u5728\u4e13\u7528\u7684 < think > \u5185\u5bb9\u5757\u4e2d\uff0c\u901a\u8fc7\u6709\u65e0\u8be6\u7ec6\u63a8\u7406\u6b65\u9aa4\u8fdb\u884c\u533a\u5206\u3002\u5177\u4f53\u800c\u8a00\uff0c\u5feb\u901f\u601d\u7ef4\u6a21\u5f0f\u523b\u610f\u4fdd\u6301 < think>\\n\\n<think > \u4e3a\u7a7a\u5185\u5bb9\u5757\uff0c\u800c\u6162\u901f\u601d\u7ef4\u6a21\u5f0f\u5219\u5728\u8be5\u533a\u5757\u660e\u786e\u5305\u542b\u9010\u6b65\u63a8\u7406\u8fc7\u7a0b\u3002\u7528\u6237\u53ef\u901a\u8fc7\u6307\u5b9a\u63a7\u5236\u6807\u7b7e\u9009\u62e9\u6a21\u5f0f\uff1a\u4f7f\u7528\u300c\/no_think\u300d\u542f\u7528\u5feb\u901f\u601d\u7ef4\u6a21\u5f0f\uff0c\u300c\/think\u300d\u542f\u7528\u6162\u901f\u601d\u7ef4\u6a21\u5f0f\u3002\u82e5\u672a\u63d0\u4f9b\u63a7\u5236\u6807\u7b7e\uff0c\u7cfb\u7edf\u9ed8\u8ba4\u91c7\u7528\u6162\u901f\u601d\u7ef4\u6a21\u5f0f\u3002<\/p>\n\n\n\n

\u6280\u672f\u62a5\u544a\uff1ahttps:\/\/github.com\/Tencent-Hunyuan\/Hunyuan-A13B\/blob\/main\/report\/Hunyuan_A13B_Technical_Report.pdf<\/p>\n\n\n\n

\u667a\u8c31\u7684 GLM-4.5 \u7cfb\u5217<\/strong><\/p>\n\n\n\n

\u4eca\u5e74 7 \u6708\u4efd\uff0c\u667a\u8c31\u53d1\u5e03\u4e86 GLM-4.5 \u548c GLM-4.5-Air<\/strong>\uff0c\u90fd\u652f\u6301\u6df7\u5408\u63a8\u7406\u6a21\u5f0f\u3002\u8be5\u6a21\u5f0f\u7684\u5f62\u6210\u4e3b\u8981\u4e0e\u6a21\u578b\u7684\u540e\u8bad\u7ec3\u6709\u5173\u3002<\/p>\n\n\n\n

\u5177\u4f53\u6765\u8bf4\uff0c\u4ed6\u4eec\u7684\u540e\u8bad\u7ec3\u5206\u4e3a\u4e24\u4e2a\u72ec\u7acb\u7684\u9636\u6bb5\u3002\u5728\u7b2c\u4e00\u9636\u6bb5\uff08\u4e13\u5bb6\u8bad\u7ec3\uff09\uff0c\u4ed6\u4eec\u6784\u5efa\u4e86\u4e13\u6ce8\u4e8e\u4e09\u4e2a\u9886\u57df\u7684\u4e13\u5bb6\u6a21\u578b\uff1a\u63a8\u7406\u3001\u4ee3\u7406\u4ee5\u53ca\u901a\u7528\u804a\u5929\u3002\u5728\u7b2c\u4e8c\u9636\u6bb5\uff08\u7edf\u4e00\u8bad\u7ec3\uff09\uff0c\u4ed6\u4eec\u91c7\u7528\u81ea\u84b8\u998f\u6280\u672f\u6765\u6574\u5408\u591a\u4e2a\u4e13\u5bb6\uff0c\u8ba9\u6a21\u578b\u5b66\u4f1a\u4e86\u4e3a\u6bcf\u4e2a\u4efb\u52a1\u5e94\u7528\u6700\u6709\u6548\u7684\u957f\u4e0a\u4e0b\u6587\u63a8\u7406\u6765\u5f97\u51fa\u51c6\u786e\u7684\u7b54\u6848\u3002\u7279\u522b\u662f\uff0c\u9274\u4e8e\u67d0\u4e9b\u9886\u57df\uff08\u5982\u95f2\u804a\uff09\u4e0d\u9700\u8981\u5197\u957f\u7684\u601d\u8003\u8fc7\u7a0b\uff0c\u4ed6\u4eec\u7cbe\u5fc3\u5e73\u8861\u4e86\u5305\u542b\u5b8c\u6574\u63a8\u7406\u8fc7\u7a0b\u7684\u8bad\u7ec3\u6570\u636e\u4e0e\u7f3a\u4e4f\u660e\u786e\u601d\u8003\u8fc7\u7a0b\u7684\u6570\u636e\u3002\u8fd9\u79cd\u65b9\u6cd5\u4f7f\u6a21\u578b\u80fd\u591f\u5728\u53cd\u601d\u548c\u5373\u65f6\u54cd\u5e94\u6a21\u5f0f\u4e4b\u95f4\u5207\u6362\uff0c\u4ece\u800c\u521b\u5efa\u4e86\u4e00\u4e2a\u6df7\u5408\u63a8\u7406\u6a21\u578b\u3002<\/p>\n\n\n\n

\u66f4\u591a\u7ec6\u8282\u53ef\u53c2\u89c1\u6280\u672f\u62a5\u544a\u3002<\/p>\n\n\n\n

\u6280\u672f\u62a5\u544a\uff1ahttps:\/\/arxiv.org\/pdf\/2508.06471<\/p>\n\n\n\n

OpenAI \u7684 GPT-5<\/strong><\/p>\n\n\n\n

\u6709\u4eba\u8bf4\uff0c\u5982\u679c GPT-3 \u5230 GPT-4 \u7684\u91cd\u5927\u7a81\u7834\u662f\u4e13\u5bb6\u6df7\u5408\uff08Mixture of Experts\uff09\uff0c\u90a3\u4e48 GPT-4o\/o3 \u5230 GPT-5 \u7684\u91cd\u5927\u7a81\u7834\u53ef\u80fd\u662f\u6a21\u578b\u6df7\u5408\uff08Mixture of Models\uff0c\u4e5f\u79f0\u4e3a\u300c\u8def\u7531\u300d\uff09\u3002<\/p>\n\n\n\n

\u548c\u5f88\u591a\u5c06\u601d\u8003 \/ \u975e\u601d\u8003\u80fd\u529b\u878d\u5408\u5230\u540c\u4e00\u4e2a\u6a21\u578b\u4e2d\u7684\u601d\u8def\u4e0d\u540c\uff0cGPT-5 \u9009\u62e9\u7684\u65b9\u5411\u662f\u5728\u6574\u4e2a\u7cfb\u7edf\u4e2d\u52a0\u5165\u4e00\u4e2a\u5b9e\u65f6\u8def\u7531\uff0c\u5b83\u80fd\u6839\u636e\u5bf9\u8bdd\u7c7b\u578b\u3001\u590d\u6742\u7a0b\u5ea6\u3001\u5de5\u5177\u9700\u6c42\u548c\u660e\u786e\u610f\u56fe\uff08\u4f8b\u5982\uff0c\u5982\u679c\u4f60\u5728\u63d0\u793a\u4e2d\u8bf4\u300c\u4ed4\u7ec6\u601d\u8003\u8fd9\u4e2a\u95ee\u9898\u300d\uff09\uff0c\u5feb\u901f\u51b3\u5b9a\u4f7f\u7528\u54ea\u4e2a\u6a21\u578b\uff08\u5982\u4e0b\u8868\uff09\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5728 GPT-5 \u6280\u672f\u62a5\u544a\u4e2d\uff0c\u4ed6\u4eec\u5c06\u5feb\u901f\u3001\u9ad8\u901a\u91cf\u7684\u6a21\u578b\u6807\u8bb0\u4e3a gpt-5-main \u548c gpt-5-main-mini\uff0c\u5c06\u601d\u8003\u578b\u6a21\u578b\u6807\u8bb0\u4e3a gpt-5-thinking \u548c gpt-5-thinking-mini\u3002API \u4e2d\u8fd8\u63d0\u4f9b\u66f4\u5c0f\u66f4\u5feb\u7684\u601d\u8003\u578b\u6a21\u578b nano \u7248\u672c\uff0cChatGPT \u4e2d\u8fd8\u63d0\u4f9b gpt-5-thinking-pro\u3002\u8fd9\u4e9b\u6a21\u578b\u5747\u7531\u4e0a\u4e00\u4ee3\u6a21\u578b\uff08\u5de6\u8fb9\u4e00\u680f\uff09\u6f14\u53d8\u800c\u6765\u3002<\/sup><\/p>\n\n\n\n

\u8be5\u8def\u7531\u901a\u8fc7\u771f\u5b9e\u4fe1\u53f7\u6301\u7eed\u8fdb\u884c\u8bad\u7ec3\uff0c\u5305\u62ec\u7528\u6237\u4f55\u65f6\u5207\u6362\u6a21\u578b\u3001\u5bf9\u56de\u590d\u7684\u504f\u597d\u4ee5\u53ca\u6d4b\u91cf\u7684\u6b63\u786e\u7387\u7b49\uff0c\u968f\u7740\u65f6\u95f4\u63a8\u79fb\u4e0d\u65ad\u6539\u8fdb\u3002\u4e00\u65e6\u8fbe\u5230\u4f7f\u7528\u9650\u5236\uff0c\u6bcf\u4e2a\u6a21\u578b\u7684\u8ff7\u4f60\u7248\u672c\u5c06\u5904\u7406\u5269\u4f59\u7684\u67e5\u8be2\u3002<\/p>\n\n\n\n

\u4e0d\u8fc7\uff0c\u8fd9\u4e2a\u6a21\u5f0f\u540c\u6837\u53cd\u54cd\u4e0d\u4f73\u3002\u4e0d\u5c11\u4eba\u5728\u793e\u4ea4\u5a92\u4f53\u4e0a\u5410\u69fd\u81ea\u5df1\u7684\u95ee\u9898\u88ab\u8def\u7531\u5230\u4e86\u4f4e\u8d28\u91cf\u6a21\u578b\u3002\u66f4\u8ba9\u4eba\u6293\u72c2\u7684\u662f\uff0c\u5f88\u591a\u65f6\u5019\u4f60\u65e0\u6cd5\u5224\u65ad\u8be5\u4e0d\u8be5\u76f8\u4fe1\u6a21\u578b\u7ed9\u51fa\u7684\u7b54\u6848\uff0c\u56e0\u4e3a\u8def\u7531\u7ed3\u679c\u662f\u4e0d\u900f\u660e\u7684\u3002\u8fd9\u8ba9 ChatGPT \u5728\u4e13\u4e1a\u7528\u6237\u4e2d\u7684\u53e3\u7891\u6709\u6240\u4e0b\u6ed1\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n

\u4e0d\u8fc7\uff0c\u5bf9\u4e8e\u5360 ChatGPT \u7528\u6237\u6570\u8d85 95% \u7684\u514d\u8d39\u7528\u6237\u6765\u8bf4\uff0c\u8fd9\u4e2a\u8def\u7531\u53cd\u800c\u63d0\u5347\u4e86\u4f53\u9a8c\u3002\u4e4b\u524d\uff0c\u8fd9\u90e8\u5206\u7528\u6237\u662f\u5f88\u96be\u7528\u4e0a\u9ad8\u7ea7\u601d\u8003\u6a21\u578b\u7684\uff0c\u4f46\u662f\u73b0\u5728\u6709\u4e00\u5b9a\u6982\u7387\u4f1a\u88ab\u8def\u7531\u5230\u9ad8\u7ea7\u6a21\u578b\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u5bf9\u6b64\uff0cSemiAnalysis CEO Dylan Patel \u5206\u6790\u8bf4\uff0c\u8fd9\u53ef\u80fd\u662f OpenAI \u5728\u514d\u8d39\u7528\u6237\u53d8\u73b0\u4e0a\u8fc8\u51fa\u7684\u91cd\u8981\u4e00\u6b65\u3002\u548c\u4e13\u653b to B \u6a21\u5f0f\u7684 Anthropic \u4e0d\u540c\uff0cOpenAI \u7684\u5546\u4e1a\u91cd\u5fc3\u4f9d\u7136\u96c6\u4e2d\u5728 C \u7aef\u7528\u6237\u4e0a\uff0c\u4f46\u8fd9\u90e8\u5206\u7528\u6237\u5927\u90e8\u5206\u662f\u514d\u8d39\u7528\u6237\u3002\u5bf9\u4e8e\u8fd9\u79cd\u60c5\u51b5\uff0c\u4f20\u7edf APP \u4e00\u822c\u662f\u901a\u8fc7\u8ba9\u514d\u8d39\u7528\u6237\u770b\u5e7f\u544a\u6765\u8d5a\u94b1\uff0c\u4f46\u5bf9\u4e8e AI \u5e94\u7528\uff0c\u8fd9\u79cd\u6a21\u5f0f\u4e0d\u518d\u9002\u7528\u3002<\/p>\n\n\n\n

\u8def\u7531\u6a21\u578b\u5b58\u5728\u7684\u4ef7\u503c\u5728\u4e8e\uff0c\u5b83\u53ef\u4ee5\u4ece\u6d77\u91cf\u514d\u8d39\u7528\u6237\u7684\u63d0\u95ee\u4e2d\u8bc6\u522b\u51fa\u5546\u4e1a\u610f\u56fe<\/strong>\uff0c\u6bd4\u5982\u8ba2\u673a\u7968\u3001\u627e\u5f8b\u5e08\uff0c\u7136\u540e\u628a\u8fd9\u4e9b\u9ad8\u4ef7\u503c\u8bf7\u6c42\u5bfc\u5411\u9ad8\u7b97\u529b\u6a21\u578b + \u540e\u7eed Agent \u670d\u52a1\uff0cOpenAI \u518d\u4ece\u6210\u4ea4\u4e2d\u62bd\u6210\u3002\u8def\u7531\u6a21\u5f0f\u8ba9 OpenAI \u7b2c\u4e00\u6b21\u628a\u300c\u6210\u672c\u300d\u548c\u300c\u5546\u4e1a\u4ef7\u503c\u300d\u5199\u8fdb\u6a21\u578b\u51b3\u7b56\u903b\u8f91\uff0c\u65e2\u7701\u7b97\u529b\uff0c\u53c8\u4e3a\u4e0b\u4e00\u6b65\u300cAI \u8d85\u7ea7\u5e94\u7528\u62bd\u6210\u300d\u94fa\u597d\u4e86\u8def\u3002<\/p>\n\n\n\n

\u4e0d\u8fc7\uff0c\u8def\u7531\u672a\u5fc5\u662f\u5b9e\u73b0\u8fd9\u4e9b\u76ee\u6807\u7684\u7ec8\u6781\u65b9\u5f0f\u3002OpenAI \u8868\u793a\uff0c\u4ed6\u4eec\u4e4b\u540e\u4e5f\u6253\u7b97\u5c06\u4e24\u79cd\u601d\u8003\u6a21\u5f0f\u7684\u5207\u6362\u6574\u5408\u5230\u5355\u4e2a\u6a21\u578b\u91cc\u3002<\/p>\n\n\n\n

DeepSeek \u7684 DeepSeek v3.1<\/strong><\/p>\n\n\n\n

DeepSeek \u6700\u8fd1\u53d1\u5e03\u7684 v3.1 \u662f\u56fd\u5185\u56e2\u961f\u5728\u300c\u5355\u4e00\u6a21\u578b\u5b9e\u73b0\u601d\u8003 \/ \u975e\u601d\u8003\u6a21\u5f0f\u5207\u6362\u300d\u4e0a\u7684\u53e6\u4e00\u9879\u5c1d\u8bd5\u3002DeepSeek \u5b98\u65b9\u8868\u793a\uff0cDeepSeek-V3.1-Think \u5b9e\u73b0\u4e86\u4e0e DeepSeek-R1-0528 \u76f8\u5f53\u7684\u7b54\u6848\u8d28\u91cf\uff0c\u540c\u65f6\u54cd\u5e94\u901f\u5ea6\u66f4\u5feb\u3002<\/p>\n\n\n\n

\u5bf9\u4e8e\u5f00\u53d1\u8005\u6765\u8bf4\uff0c\u5b83\u7684\u601d\u8003\u6a21\u5f0f\u548c\u975e\u601d\u8003\u6a21\u5f0f\u53ef\u4ee5\u7531\u63d0\u793a\u5e8f\u5217\u4e2d\u7684 <think> \u548c <\/think> \u6807\u8bb0\u89e6\u53d1\u3002\u5bf9\u4e8e C \u7aef\u7528\u6237\uff0c\u53ef\u4ee5\u901a\u8fc7\u70b9\u51fb\u300c\u6df1\u5ea6\u601d\u8003\u300d\u6309\u94ae\u5207\u6362\u6a21\u5f0f\u3002<\/p>\n\n\n\n

\u7531\u4e8e\u53d1\u5e03\u65f6\u95f4\u63a5\u8fd1\uff0c\u53c8\u90fd\u6709\u6df7\u5408\u63a8\u7406\u6a21\u5f0f\uff0cDeepSeek v3.1 \u548c GPT-5 \u96be\u514d\u88ab\u62ff\u6765\u5bf9\u6bd4\u3002\u5728\u6027\u80fd\u4e0a\uff0cDeepSeek v3.1 \u867d\u7136\u5728\u4e00\u4e9b\u57fa\u51c6\u4e0a\u4e0e GPT-5 \u65d7\u9f13\u76f8\u5f53\uff0c\u4f46\u7efc\u5408\u80fd\u529b\u4ecd\u7136\u4e0d\u5982 GPT-5\u3002\u5728\u4ef7\u683c\u4e0a\uff0cDeepSeek v3.1 \u5219\u6709\u7740\u660e\u663e\u7684\u4f18\u52bf\uff0c\u53ef\u4ee5\u8bf4\u4e3a\u4f01\u4e1a\u63d0\u4f9b\u4e86\u4e00\u4e2a\u9ad8\u6027\u4ef7\u6bd4\u7684\u5f00\u6e90\u9009\u62e9\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u60f3\u6df1\u5165\u4e86\u89e3\u6df7\u5408\u63a8\u7406\uff1f \u8fd9\u4e9b\u7814\u7a76\u65b9\u5411\u503c\u5f97\u5173\u6ce8<\/strong><\/p>\n\n\n\n

\u4ece\u4ee5\u4e0a\u6a21\u578b\u53ef\u4ee5\u770b\u51fa\uff0c\u867d\u7136\u5927\u5bb6\u7684\u5171\u540c\u76ee\u6807\u90fd\u662f\u51cf\u5c11\u63a8\u7406\u8fc7\u7a0b\u4e2d\u7684 token \u6d6a\u8d39\uff0c\u4f46\u5177\u4f53\u5b9e\u73b0\u65b9\u6cd5\u6709\u6240\u4e0d\u540c\uff0c\u6709\u7684\u501f\u7528\u8def\u7531\u5c06\u95ee\u9898\u5bfc\u5411\u4e0d\u540c\u7684\u6a21\u578b\uff0c\u8fd8\u6709\u4e9b\u5728\u4e00\u4e2a\u6a21\u578b\u4e2d\u5b9e\u73b0\u5feb\u6162\u601d\u8003\u7684\u5207\u6362\u3002\u5728\u5207\u6362\u65b9\u5f0f\u4e0a\uff0c\u6709\u4e9b\u662f\u7528\u6237\u663e\u5f0f\u63a7\u5236\uff0c\u6709\u4e9b\u662f\u6a21\u578b\u81ea\u52a8\u5224\u65ad\u3002<\/p>\n\n\n\n

\u901a\u8fc7\u4e00\u4e9b\u7efc\u8ff0\u7814\u7a76\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u5230\u66f4\u591a\u4e0d\u540c\u7684\u601d\u8def\u3002<\/p>\n\n\n\n

\u6bd4\u5982\u5728\u300cTowards Concise and Adaptive Thinking in Large Reasoning Models: A Survey\u300d\u8fd9\u7bc7\u7efc\u8ff0\u4e2d\uff0c\u7814\u7a76\u8005\u5c06\u73b0\u6709\u65b9\u6cd5\u5206\u4e3a\u4e24\u7c7b\uff1a<\/p>\n\n\n\n

\u4e00\u7c7b\u662f\u65e0\u9700\u8bad\u7ec3\u7684\u65b9\u6cd5<\/strong>\uff0c\u5305\u62ec\u63d0\u793a\u8bcd\u5f15\u5bfc\u3001\u57fa\u4e8e pipeline \u7684\u65b9\u6cd5\uff08\u6bd4\u5982\u8def\u7531\uff09\u3001\u89e3\u7801\u64cd\u7eb5\u548c\u6a21\u578b\u878d\u5408\u7b49\uff1b<\/p>\n\n\n\n

    \n
  • \u63d0\u793a\u8bcd\u5f15\u5bfc<\/strong>\uff1a\u901a\u8fc7\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u63d0\u793a\uff08\u4f8b\u5982\uff0c\u76f4\u63a5\u63d0\u793a\u3001token \u9884\u7b97\u3001thinking \u6a21\u5f0f\u3001no-thinking \u6307\u4ee4\uff09\u6765\u5229\u7528\u6a21\u578b\u9075\u5faa\u6307\u4ee4\u7684\u80fd\u529b\u3002\u5c3d\u7ba1\u8be5\u65b9\u6cd5\u7684\u7b80\u5355\u6027\u4f7f\u5176\u80fd\u591f\u5feb\u901f\u90e8\u7f72\uff0c\u4f46\u5176\u6709\u6548\u6027\u53d6\u51b3\u4e8e\u6a21\u578b\u5bf9\u7ea6\u675f\u7684\u9075\u5b88\u60c5\u51b5\uff0c\u800c\u8fd9\u5f80\u5f80\u5e76\u4e0d\u4e00\u81f4\u3002\u7814\u7a76\u8868\u660e\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u4f1a\u4ea7\u751f\u610f\u60f3\u4e0d\u5230\u7684\u540e\u679c\uff0c\u4f8b\u5982\u9690\u85cf\u7684\u4e0d\u51c6\u786e\u4e4b\u5904\u548c\u8f93\u51fa\u7a33\u5b9a\u6027\u7684\u964d\u4f4e\uff0c\u7279\u522b\u662f\u5728\u5b9e\u65bd\u4e25\u683c\u7684 token \u9650\u5236\u6216\u6291\u5236\u63a8\u7406\u6b65\u9aa4\u65f6\u3002<\/li>\n\n\n\n
  • \u57fa\u4e8e pipeline \u7684\u65b9\u6cd5<\/strong>\uff1a\u8be5\u65b9\u6cd5\u5c06\u63a8\u7406\u5de5\u4f5c\u6d41\u7a0b\u6a21\u5757\u5316\uff0c\u901a\u8fc7\u4efb\u52a1\u5378\u8f7d\u964d\u4f4e\u5927\u8bed\u8a00\u63a8\u7406\u6a21\u578b\u7684\u8ba1\u7b97\u6210\u672c\uff0c\u540c\u65f6\u4fdd\u6301\u63a8\u7406\u8d28\u91cf\u3002\u5176\u4e2d\uff0c\u57fa\u4e8e\u8def\u7531\u7684\u65b9\u6cd5\u6839\u636e\u8f93\u5165\u590d\u6742\u6027\u3001\u6a21\u578b\u80fd\u529b\u6216\u9884\u7b97\u9650\u5236\u52a8\u6001\u9009\u62e9\u6700\u4f73\u6a21\u578b \/ \u63a8\u7406\u6a21\u5f0f\u3002\u5176\u4ed6\u7b56\u7565\u5305\u62ec\u52a8\u6001\u89c4\u5212\u548c\u8fed\u4ee3\u4f18\u5316\u4ee5\u53ca\u6548\u7387\u63d0\u5347\u6280\u672f\u3002\u8fd9\u4e9b\u65b9\u6cd5\u663e\u8457\u7f29\u77ed\u4e86\u63a8\u7406\u957f\u5ea6\uff0c\u4f46\u5f15\u5165\u4e86\u989d\u5916\u7684\u5f00\u9500\uff08\u5982\u8def\u7531\u5ef6\u8fdf\uff09\uff0c\u5bfc\u81f4\u7aef\u5230\u7aef\u5ef6\u8fdf\u589e\u52a0\uff0c\u56e0\u6b64\u9700\u8981\u5728\u6548\u7387\u548c\u5ef6\u8fdf\u4e4b\u95f4\u8fdb\u884c\u6743\u8861\u3002<\/li>\n\n\n\n
  • \u89e3\u7801\u64cd\u7eb5<\/strong>\uff1a\u901a\u8fc7\u9884\u7b97\u5f3a\u5236\u3001\u63d0\u524d\u9000\u51fa\u68c0\u67e5\u3001logit \u8c03\u6574\u6216\u6fc0\u6d3b\u5f15\u5bfc\u7b49\u65b9\u5f0f\uff0c\u52a8\u6001\u4ecb\u5165\u751f\u6210\u8fc7\u7a0b\u3002\u50cf DEER \u548c FlashThink \u8fd9\u7c7b\u6280\u672f\uff0c\u901a\u8fc7\u76d1\u6d4b\u7f6e\u4fe1\u5ea6\u6216\u8bed\u4e49\u6536\u655b\u6765\u5b9e\u73b0\u66f4\u77ed\u7684\u63a8\u7406\u94fe\uff0c\u4e0d\u8fc7\u9891\u7e41\u7684\u9a8c\u8bc1\u6b65\u9aa4\u53ef\u80fd\u4f1a\u62b5\u6d88\u8ba1\u7b97\u8282\u7701\u3002\u5e76\u884c scaling \u7b56\u7565\u8fdb\u4e00\u6b65\u63d0\u9ad8\u4e86\u6548\u7387\uff0c\u4f46\u9700\u8981\u4ed4\u7ec6\u6821\u51c6\u4ee5\u5e73\u8861\u5197\u4f59\u5ea6\u548c\u51c6\u786e\u6027\u3002<\/li>\n\n\n\n
  • \u6a21\u578b\u878d\u5408<\/strong>\uff1a\u5373\u5c06\u4e00\u4e2a\u601d\u8003\u7f13\u6162\u7684\u5927\u8bed\u8a00\u63a8\u7406\u6a21\u578b\uff08LRM\uff09\u548c\u4e00\u4e2a\u601d\u8003\u5feb\u901f\u7684\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u6574\u5408\u4e3a\u4e00\u4e2a\u5355\u4e00\u6a21\u578b\uff0c\u5e76\u4e14\u671f\u671b\u8fd9\u4e2a\u5355\u4e00\u6a21\u578b\u80fd\u591f\u5e73\u8861\u5feb\u6162\u601d\u8003\uff0c\u4ece\u800c\u5b9e\u73b0\u81ea\u9002\u5e94\u601d\u8003\u3002\u8fd9\u79cd\u65b9\u6cd5\u901a\u8fc7\u53c2\u6570\u63d2\u503c\u6216\u57fa\u4e8e\u6fc0\u6d3b\u7684\u878d\u5408\u6765\u7efc\u5408\u957f\u63a8\u7406\u548c\u77ed\u63a8\u7406\u80fd\u529b\u3002\u867d\u7136\u8fd9\u79cd\u65b9\u6cd5\u5bf9\u4e2d\u7b49\u89c4\u6a21\u7684\u6a21\u578b\u6709\u6548\uff0c\u4f46\u5728\u5904\u7406\u6781\u7aef\u89c4\u6a21\uff08\u5c0f\u578b\u6216\u5927\u578b\u6a21\u578b\uff09\u65f6\u5b58\u5728\u56f0\u96be\uff0c\u5e76\u4e14\u7f3a\u4e4f\u5bf9\u63a8\u7406\u6df1\u5ea6\u7684\u7cbe\u7ec6\u63a7\u5236\u3002\u4e0e\u6b64\u540c\u65f6\uff0c\u50cf Activation-Guided Consensus Merging (ACM) \u8fd9\u6837\u7684\u6700\u65b0\u8fdb\u5c55\u51f8\u663e\u4e86\u4e92\u4fe1\u606f\u5206\u6790\u5728\u5bf9\u9f50\u5f02\u6784\u6a21\u578b\u65b9\u9762\u7684\u6f5c\u529b\u3002\u00a0<\/li>\n<\/ul>\n\n\n\n

    \u53e6\u4e00\u7c7b\u662f\u57fa\u4e8e\u8bad\u7ec3\u7684\u65b9\u6cd5<\/strong>\uff0c\u91cd\u70b9\u5728\u4e8e\u7f29\u77ed\u63a8\u7406\u957f\u5ea6\uff0c\u5e76\u901a\u8fc7\u5fae\u8c03\uff08SFT\/DPO\uff09\u6216\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u6765\u6559\u5bfc\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u81ea\u9002\u5e94\u601d\u8003\u3002<\/p>\n\n\n\n

      \n
    • \u5fae\u8c03<\/strong>\uff1a\u5fae\u8c03\u53ef\u4ee5\u5206\u4e3a\u4e94\u7c7b\uff1a\u957f\u601d\u7ef4\u94fe\u538b\u7f29<\/strong>\u65b9\u6cd5\u63d0\u9ad8\u4e86\u63a8\u7406\u6548\u7387\u548c\u9002\u5e94\u6027\uff0c\u4f46\u5728\u538b\u7f29\u6548\u679c\u4e0e\u63a8\u7406\u4fdd\u771f\u5ea6\u4e4b\u95f4\u9762\u4e34\u6743\u8861\uff0c\u540c\u65f6\u8fd8\u5b58\u5728\u6570\u636e\u9700\u6c42\u589e\u52a0\u548c\u6cdb\u5316\u65b9\u9762\u7684\u6311\u6218\uff1b\u800c\u77ed\u601d\u7ef4\u94fe\u9009\u62e9\u5fae\u8c03<\/strong>\u5219\u901a\u8fc7\u4fc3\u8fdb\u7b80\u6d01\u6216\u81ea\u6211\u9a8c\u8bc1\u7684\u63a8\u7406\u8def\u5f84\u6765\u63d0\u9ad8\u63a8\u7406\u6548\u7387\uff0c\u4f46\u53ef\u80fd\u5b58\u5728\u9057\u6f0f\u5173\u952e\u6b65\u9aa4\u7684\u98ce\u9669\uff0c\u6216\u8005\u9700\u8981\u590d\u6742\u7684\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u5e76\u5728\u7b80\u6d01\u6027\u548c\u51c6\u786e\u6027\u4e4b\u95f4\u8fdb\u884c\u4ed4\u7ec6\u6743\u8861\uff1b\u9690\u5f0f\u601d\u7ef4\u94fe\u5fae\u8c03<\/strong>\u901a\u8fc7\u6f5c\u5728\u63a8\u7406\u8868\u793a\u6216\u77e5\u8bc6\u84b8\u998f\u6765\u5b9e\u73b0\u6548\u7387\u63d0\u5347\uff0c\u4f46\u7531\u4e8e\u63a8\u7406\u6b65\u9aa4\u4e0d\u660e\u786e\u800c\u727a\u7272\u4e86\u89e3\u91ca\u6027\uff0c\u4e14\u538b\u7f29\u8868\u793a\u4e0e\u4efb\u52a1\u8981\u6c42\u4e4b\u95f4\u53ef\u80fd\u5b58\u5728\u4e0d\u4e00\u81f4\u7684\u98ce\u9669\uff1b\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\uff08DPO\uff09\u53d8\u4f53<\/strong>\u65b9\u6cd5\u901a\u8fc7\u504f\u597d\u5b66\u4e60\u5b9e\u73b0\u7b80\u6d01\u6027\u548c\u51c6\u786e\u6027\u4e4b\u95f4\u7684\u591a\u76ee\u6807\u4f18\u5316\u5e73\u8861\uff0c\u4f46\u5728\u6784\u5efa\u9ad8\u8d28\u91cf\u504f\u597d\u5bf9\u4ee5\u53ca\u5728\u4e25\u683c\u957f\u5ea6\u9650\u5236\u4e0b\u4fdd\u6301\u63a8\u7406\u6df1\u5ea6\u65b9\u9762\u9762\u4e34\u6311\u6218\uff1b\u5176\u4ed6\u6df7\u5408\u65b9\u6cd5<\/strong>\u7ed3\u5408\u4e86\u5feb\u901f \/ \u6162\u901f\u8ba4\u77e5\u7cfb\u7edf\u6216\u65b0\u9896\u7684\u635f\u5931\u51fd\u6570\u6765\u5b9e\u73b0\u81ea\u9002\u5e94\u63a8\u7406\uff0c\u4e0d\u8fc7\u5b83\u4eec\u901a\u5e38\u9700\u8981\u590d\u6742\u7684\u8bad\u7ec3\u6d41\u7a0b\uff0c\u5e76\u5bf9\u53cc\u6a21\u5f0f\u4ea4\u4e92\u8fdb\u884c\u4ed4\u7ec6\u6821\u51c6\u3002<\/li>\n\n\n\n
    • \u5f3a\u5316\u5b66\u4e60<\/strong>\uff1a\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5\u901a\u8fc7\u4e94\u4e2a\u5173\u952e\u8303\u5f0f\u6765\u5e73\u8861\u7b80\u6d01\u6027\u548c\u51c6\u786e\u6027\u3002\u5e26\u957f\u5ea6\u60e9\u7f5a\u7684\u5f3a\u5316\u5b66\u4e60<\/strong>\u901a\u8fc7\u5956\u52b1\u5851\u9020\u6216\u5916\u90e8\u7ea6\u675f\u5bf9\u5197\u957f\u7684\u8f93\u51fa\u8fdb\u884c\u60e9\u7f5a\uff0c\u4ece\u800c\u63d0\u9ad8\u6548\u7387\uff0c\u4f46\u5b58\u5728\u5c06\u590d\u6742\u4efb\u52a1\u8fc7\u5ea6\u7b80\u5316\u6216\u8fc7\u5ea6\u62df\u5408\u60e9\u7f5a\u9608\u503c\u7684\u98ce\u9669\u3002GRPO \u53d8\u4f53\u65b9\u6cd5<\/strong>\u901a\u8fc7\u4f7f\u63a8\u7406\u6a21\u5f0f\u591a\u6837\u5316\u6216\u6574\u5408\u96be\u5ea6\u611f\u77e5\u5956\u52b1\u6765\u89e3\u51b3\u300c\u683c\u5f0f\u5d29\u6e83\u300d\u95ee\u9898\uff0c\u4e0d\u8fc7\u5b83\u4eec\u901a\u5e38\u9700\u8981\u590d\u6742\u7684\u635f\u5931\u8bbe\u8ba1\u548c\u591a\u7ec4\u4ef6\u7cfb\u7edf\u3002\u96be\u5ea6\u611f\u77e5\u5f3a\u5316\u5b66\u4e60<\/strong>\u901a\u8fc7\u663e\u5f0f\u96be\u5ea6\u4f30\u8ba1\u6216\u9690\u5f0f\u4fe1\u53f7\uff08\u54cd\u5e94\u957f\u5ea6\u3001\u89e3\u51b3\u7387\uff09\u4f7f\u54cd\u5e94\u957f\u5ea6\u9002\u5e94\u95ee\u9898\u7684\u590d\u6742\u6027\uff0c\u4f46\u5728\u51c6\u786e\u7684\u96be\u5ea6\u6821\u51c6\u548c\u8de8\u9886\u57df\u6cdb\u5316\u65b9\u9762\u9762\u4e34\u6311\u6218\u3002\u601d\u7ef4\u6a21\u5f0f\u5f3a\u5316\u5b66\u4e60<\/strong>\u80fd\u591f\u5728\u5ba1\u614e\uff08\u300c\u601d\u8003\u300d\uff09\u548c\u53cd\u5e94\u6027\uff08\u300c\u4e0d\u601d\u8003\u300d\uff09\u6a21\u5f0f\u4e4b\u95f4\u52a8\u6001\u5207\u6362\uff0c\u4f46\u5728\u6a21\u5f0f\u9009\u62e9\u7a33\u5b9a\u6027\u548c\u63a2\u7d22\u4e0e\u5229\u7528\u7684\u6743\u8861\u65b9\u9762\u5b58\u5728\u56f0\u96be\u3002\u5176\u4ed6\u5f3a\u5316\u5b66\u4e60\u521b\u65b0<\/strong>\u5f15\u5165\u4e86\u53ef\u5b66\u4e60\u7684\u5956\u52b1\u51fd\u6570\u3001\u6df7\u5408\u6846\u67b6\u6216\u65b0\u9896\u7684\u6307\u6807\uff0c\u5c3d\u7ba1\u8fd9\u4e9b\u901a\u5e38\u9700\u8981\u5927\u91cf\u7684\u8ba1\u7b97\u8d44\u6e90\u6216\u9762\u4e34\u53ef\u6269\u5c55\u6027\u95ee\u9898\u3002<\/li>\n<\/ul>\n\n\n\n

      \u5177\u4f53\u5206\u7c7b\u5982\u4e0b\u56fe\u6240\u793a\uff1a<\/p>\n\n\n\n

      \"\"<\/figure>\n\n\n\n

      \u7efc\u8ff0\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2507.09662<\/p>\n\n\n\n

      \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u9664\u4e86\u8bed\u8a00\u6a21\u578b\uff0c\u591a\u6a21\u6001\u6a21\u578b\u9886\u57df\u7684\u6df7\u5408\u63a8\u7406\u63a2\u7d22\u4e5f\u5df2\u7ecf\u5f00\u59cb\uff0c\u800c\u4e14\u51fa\u73b0\u4e86 R-4B \u7b49\u81ea\u52a8\u5316\u7a0b\u5ea6\u8f83\u9ad8\u7684\u81ea\u9002\u5e94\u601d\u8003\u6a21\u578b\uff0c\u6211\u4eec\u5c06\u5728\u540e\u7eed\u7684\u62a5\u9053\u4e2d\u5b8c\u6574\u5448\u73b0\u3002<\/p>\n\n\n\n

      \u5982\u679c\u4f60\u60f3\u52a8\u6001\u8ffd\u8e2a\u8fd9\u4e2a\u9886\u57df\u7684\u65b0\u7814\u7a76\uff0c\u53ef\u4ee5\u6536\u85cf\u4ee5\u4e0b GitHub \u9879\u76ee\uff1ahttps:\/\/github.com\/hemingkx\/Awesome-Efficient-Reasoning#adaptive-thinking<\/p>\n\n\n\n

      \"\"<\/figure>\n\n\n\n

      \u4e0b\u4e00\u4e2a\u524d\u6cbf\uff1a\u8ba9 AI \u4ee5\u6700\u4f4e\u4ee3\u4ef7\u5728\u6070\u5f53\u65f6\u523b\u601d\u8003<\/strong><\/p>\n\n\n\n

      \u5728\u8fc7\u53bb\u51e0\u5e74\uff0cAI \u9886\u57df\u7684\u7ade\u4e89\u66f4\u591a\u96c6\u4e2d\u5728\u6784\u5efa\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u4e0a\u3002\u5982\u4eca\uff0c\u6df7\u5408\u63a8\u7406\u6a21\u5f0f\u7684\u5927\u89c4\u6a21\u51fa\u73b0\u6807\u5fd7\u7740\u4eba\u5de5\u667a\u80fd\u884c\u4e1a\u7684\u91cd\u70b9\u4ece\u5355\u7eaf\u6784\u5efa\u66f4\u5f3a\u5927\u7684\u7cfb\u7edf\u8f6c\u5411\u521b\u5efa\u5b9e\u7528\u7684\u7cfb\u7edf\u3002\u6b63\u5982 IBM \u7814\u7a76\u9662\u9ad8\u7ea7\u9879\u76ee\u7ecf\u7406 Abraham Daniels \u6240\u8bf4\uff0c\u5bf9\u4e8e\u4f01\u4e1a\u800c\u8a00\uff0c\u8fd9\u79cd\u8f6c\u53d8\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u8fd0\u8425\u590d\u6742\u4eba\u5de5\u667a\u80fd\u7684\u6210\u672c\u5df2\u6210\u4e3a\u4e3b\u8981\u8003\u8651\u56e0\u7d20\u3002<\/p>\n\n\n\n

      \u4f46\u662f\uff0c\u8fd9\u4e00\u8f6c\u53d8\u4e5f\u5728\u7ecf\u5386\u9635\u75db\u3002\u4e00\u65b9\u9762\uff0c\u80fd\u591f\u4e0d\u9760\u4eba\u7c7b\u6307\u793a\u6fc0\u6d3b\u6df1\u5ea6\u601d\u8003\u6a21\u5f0f\u7684\u6210\u529f\u6a21\u578b\u8fd8\u76f8\u5bf9\u8f83\u5c11\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5c1d\u8bd5\u53bb\u6389\u663e\u5f0f\u5f00\u5173\u7684\u601d\u7ef4\u8f6c\u6362\u6a21\u5f0f\u8fd8\u6ca1\u6709\u5b9e\u73b0\u8db3\u591f\u4ee4\u4eba\u6ee1\u610f\u7684\u6548\u679c\u3002\u8fd9\u4e9b\u73b0\u8c61\u90fd\u8bf4\u660e\uff0c\u6df7\u5408\u63a8\u7406\u7684\u4e0b\u4e00\u4e2a\u524d\u6cbf\u5c06\u662f\u66f4\u667a\u80fd\u7684\u81ea\u6211\u8c03\u8282\u3002<\/p>\n\n\n\n

      \u6362\u53e5\u8bdd\u8bf4\uff0c\u6df7\u5408\u63a8\u7406\u7684\u672a\u6765\u7ade\u4e89\u5c06\u4e0d\u518d\u53ea\u662f\u300c\u662f\u5426\u80fd\u601d\u8003\u300d\uff0c\u800c\u662f\u300c\u80fd\u5426\u4ee5\u6700\u4f4e\u4ee3\u4ef7\u5728\u6070\u5f53\u65f6\u523b\u601d\u8003\u300d\u3002\u8c01\u80fd\u5728\u8fd9\u4e00\u70b9\u4e0a\u627e\u5230\u6700\u4f18\u89e3\uff0c\u8c01\u5c31\u80fd\u5728\u4e0b\u4e00\u8f6e AI \u6027\u80fd\u4e0e\u6210\u672c\u535a\u5f08\u4e2d\u5360\u636e\u4e3b\u52a8\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

      \u5728\u6700\u8fd1\u7684\u4e00\u6863\u8131\u53e3\u79c0\u8282\u76ee\u4e2d\uff0c\u6f14\u5458\u5f20\u4fca\u8c03\u4f83 DeepSeek \u662f\u4e00\u6b3e\u975e\u5e38\u300c\u5185\u8017\u300d\u7684 AI\uff0c\u8fde\u4e2a\u300c1 \u52a0 1 \u7b49\u4e8e\u51e0 […]<\/p>\n","protected":false},"author":1,"featured_media":1225,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[],"class_list":["post-1210","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-10"],"_links":{"self":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1210","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/comments?post=1210"}],"version-history":[{"count":1,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1210\/revisions"}],"predecessor-version":[{"id":1226,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/1210\/revisions\/1226"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media\/1225"}],"wp:attachment":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media?parent=1210"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/categories?post=1210"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/tags?post=1210"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}