AI“以小博大”新标杆：三星开源TRM模型，挑战谷歌与OpenAI

IT之家2025-10-10

IT之家 10 月 10 日消息，科技媒体 venturebeat 于 10 月 8 日发布博文，报道称三星高级 AI 研究院发布了名为微型递归模型（TRM）的开源 AI 模型，仅包含 700 万个参数，不过在数独、迷宫等特定的结构化推理任务上，表现媲美甚至超越了参数量为其 10000 倍的谷歌 Gemini 2.5 Pro 等顶尖大模型。该 AI 模型由三星高级 AI 研究院（SAIT）高级 ...

网页链接

免责声明：本文观点仅代表作者个人观点，不构成本平台的投资建议，本平台不对文章信息准确性、完整性和及时性做出任何保证，亦不对因使用或信赖文章信息引发的任何损失承担责任。

精彩评论

我们需要你的真知灼见来填补这片空白

发表看法

{"i18n":{"language":"zh_CN"},"isChannel":false,"data":{"share":"https://www.laohu8.com/m/news/2574147184?lang=zh_CN&edition=full","thumbnail":"","is_english":false,"pubTime":"2025-10-10 14:40","share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","id":"2574147184","market":"us","top_or_hot":-1,"title":"AI“以小博大”新标杆：三星开源TRM模型，挑战谷歌与OpenAI","media":"IT之家","content":"<div>\n<p>IT之家 10 月 10 日消息，科技媒体 venturebeat 于 10 月 8 日发布博文，报道称三星高级 AI 研究院发布了名为微型递归模型（TRM）的开源 AI 模型，仅包含 700 万个参数，不过在数独、迷宫等特定的结构化推理任务上，表现媲美甚至超越了参数量为其 10000 倍的谷歌 Gemini 2.5 Pro 等顶尖大模型。该 AI 模型由三星高级 AI 研究院（SAIT）高级 ...</p>\n\n<a href=\"https://tech.ifeng.com/c/8nL0pJNJagz\">网页链接</a>\n\n</div>\n","source":"fenghuang_stock","html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>AI“以小博大”新标杆：三星开源TRM模型，挑战谷歌与OpenAI</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\nAI“以小博大”新标杆：三星开源TRM模型，挑战谷歌与OpenAI\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-10-10 14:40 北京时间&nbsp;&nbsp;&nbsp;<a href=https://tech.ifeng.com/c/8nL0pJNJagz><strong>IT之家</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>IT之家 10 月 10 日消息，科技媒体 venturebeat 于 10 月 8 日发布博文，报道称三星高级 AI 研究院发布了名为微型递归模型（TRM）的开源 AI 模型，仅包含 700 万个参数，不过在数独、迷宫等特定的结构化推理任务上，表现媲美甚至超越了参数量为其 10000 倍的谷歌 Gemini 2.5 Pro 等顶尖大模型。该 AI 模型由三星高级 AI 研究院（SAIT）高级 ...</p>\n\n<a href=\"https://tech.ifeng.com/c/8nL0pJNJagz\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"LU1935043023.USD","symbol_name":"MANULIFE GF GLOBAL MULTI-ASSET DIVERSIFIED INCOME \"AA\" (USD) INC A","start_time":0,"source_url":"https://tech.ifeng.com/c/8nL0pJNJagz","article_id":"2574147184","we_media_id":null,"thumbnails":[],"rights":null,"url":"https://stock-news.laohu8.com/highlight/detail?id=2574147184","pubTimestamp":1760078415,"columns":[],"sourceInfo":{"source_id":"fenghuang_stock","name":"凤凰网"},"weMediaInfo":null,"summary":"IT之家 10 月 10 日消息，科技媒体 venturebeat 于 10 月 8 日发布博文，报道称三星高级 AI 研究院发布了名为微型递归模型的开源 AI 模型，仅包含 700 万个参数，不过在数独、迷宫等特定的结构化推理任务上，表现媲美甚至超越了参数量为其 10000 倍的谷歌 Gemini 2.5 Pro 等顶尖大模型。TRM 的代码、训练脚本和数据集目前已在 GitHub 上根据 MIT 许可证完全开源，企业和研究人员均可免费使用、修改和部署，甚至用于商业应用。","collect":0,"end_time":0,"defaultTopTitle":"ifeng.com","property":[],"viewcount":null,"language":"zh","relate_stocks":{"LU1935043023.USD":"MANULIFE GF GLOBAL MULTI-ASSET DIVERSIFIED INCOME \"AA\" (USD) INC A","LU0215105999.USD":"SCHRODER ISF GLOBAL EQUITY \"A\" ACC","LU0107464264.USD":"abrdn SICAV I - GLOBAL INNOVATION EQUITY \"A\" (USD) ACC","BK4507":"流媒体概念","LU2471134796.USD":"INVESCO GLOBAL EQUITY INCOME ADVANTAGE \"A\" (USD) INC","LU2417539215.USD":"ALLIANZ GLOBAL INCOME \"AMF\" (USD) INC","LU0187121727.USD":"FIDELITY SUSTAINABLE US EQUITY \"A\" (USD) INC","LU2458330243.SGD":"FRANKLIN SHARIAH TECHNOLOGY \"A-H1\" (SGDHDG) ACC","LU0820562030.AUD":"ALLIANZ INCOME AND GROWTH \"AMH2\" (AUDHDG) H2 INC","LU2750360641.GBP":"INVESCO GLOBAL EQUITY INCOME ADVANTAGE \"A\" (GBPHDG) INC","LU1778281490.HKD":"HSBC GIF GLOBAL LOWER CARBON EQUITY \"AD\" (HKD) INC","IE00BKPKM429.USD":"NEUBERGER BERMAN GLOBAL SUSTAINABLE EQUITY \"A\" (USD) ACC","SGXZ31699556.SGD":"UGDP UNITED GLOBAL QUALITY GROWTH \"C\" (SGDHDG) ACC","LU0251132253.USD":"FIDELITY FUNDS  GLOBAL THEMATIC OPPORTUNITIES \"A\" (USD) ACC","LU0354030511.USD":"ALLSPRING  U.S. LARGE CAP GROWTH \"I\" (USD) ACC","LU2750360997.AUD":"INVESCO GLOBAL EQUITY INCOME ADVANTAGE \"A\" (AUDHDG) INC","LU0823434740.USD":"BNP PARIBAS US GROWTH \"C\" (USD) INC","LU0792757196.USD":"TEMPLETON SHARIAH GLOBAL EQUITY FUND \"A\" (USD) ACC","LU1035773651.USD":"AB SICAV I - GLOBAL VALUE PORTFOLIO \"AD\" (USD) INC","BK4573":"虚拟现实","LU1670628061.USD":"M&G (LUX) NORTH AMERICAN DIVIDEND \"A\" (USD) INC","LU2213496289.HKD":"ALLIANZ INCOME AND GROWTH \"AT\" (HKD) ACC","LU0070302665.USD":"FRANKLIN MUTUAL U.S. VALUE \"A\" (USD) ACC","LU2430703251.USD":"WELLINGTON MULTI-ASSET HIGH INCOME \"AM4\" (USD) INC","LU1035775433.USD":"AB SICAV I - AMERICAN GROWTH PORTFOLIO \"AD\" (USD) INC","GB00B4QBRK32.GBP":"FUNDSMITH EQUITY \"R\" (GBP) INC","LU2237443622.USD":"Aberdeen Standard SICAV I - Global Dynamic Dividend A Acc USD","BK4585":"ETF&股票定投概念","LU2065170008.USD":"M&G (LUX) GLOBAL MAXIMA \"A\" (USD) INC","LU2237443978.SGD":"Aberdeen Standard SICAV I - Global Dynamic Dividend A Acc SGD-H","LU0788109477.HKD":"BGF GLOBAL ALLOCATION \"A2\" (HKDHGD) ACC","LU0889566641.SGD":"FTSF - Templeton Shariah Global Equity A Acc SGD","LU0444971666.USD":"天利全球科技基金","LU0061475181.USD":"THREADNEEDLE (LUX) AMERICAN \"AU\" (USD) ACC","LU0109391861.USD":"富兰克林美国机遇基金A Acc","BK4587":"ChatGPT概念","IE0004091025.USD":"BNY MELLON GLOBAL OPPORTUNITIES \"B\" (USD) ACC","LU1670627923.USD":"M&G (LUX) NORTH AMERICAN DIVIDEND \"A\" (USD) ACC","BK4612":"AI芯片","LU0942090050.USD":"UBS (LUX) EQUITY SICAV - US TOTAL YIELD SUSTAINABLE \"P\" (USD)  INC","LU1366192091.USD":"ALLIANZ US EQUITY PLUS \"AM\" (USD) INC","USJW.SI":"ALPHAB 3xLongSG261006","LU0312595415.SGD":"Schroder ISF Global Climate Change Equity A Acc SGD","LU2242649171.HKD":"FIDELITY FUNDS GLOBAL THEMATIC OPPORTUNITIES \"A\" (HKD) ACC","LU0211328371.USD":"TEMPLETON GLOBAL EQUITY INCOME \"A\" (MDIS) (USD)   INC","LU0545039389.USD":"BGF GLOBAL EQUITY INCOME \"A2\" ACC","LU2054465674.USD":"UBS (LUX) KEY SELEC SICAV DIGITAL TRANSFORMATION T \"P\" (USD) ACC","IE00B1XK9C88.USD":"PINEBRIDGE US LARGE CAP RESEARCH ENHANCED \"A\" (USD) ACC"},"translate_title":"A new benchmark for AI to \"use small to broaden big\": Samsung's open source TRM model challenges Google and OpenAI","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"USJW.SI":0.6},"content_text":"IT之家 10 月 10 日消息，科技媒体 venturebeat 于 10 月 8 日发布博文，报道称三星高级 AI 研究院发布了名为微型递归模型（TRM）的开源 AI 模型，仅包含 700 万个参数，不过在数独、迷宫等特定的结构化推理任务上，表现媲美甚至超越了参数量为其 10000 倍的谷歌 Gemini 2.5 Pro 等顶尖大模型。该 AI 模型由三星高级 AI 研究院（SAIT）高级 AI 研究员 Alexia Jolicoeur-Martineau 发布，成为 AI 领域“小模型”对抗“大模型”的趋势的重磅新案例。该模型仅有 700 万参数，设计理念是极致简化复杂性。在架构方面，摒弃了分层推理模型（HRM）所依赖的双网络协作架构，转而采用一个仅有两层的单一模型。其核心机制在于“递归推理”：模型对自身输出的预测进行反复迭代和修正，每一步都纠正前一步的潜在错误，直至答案收敛稳定。TRM 通过这种方式，用迭代计算的深度模拟了庞大网络的复杂推理过程，实现了“以递归替代规模”的目标，从而在不牺牲性能的前提下，大幅降低了计算和内存成本。尽管模型极小，TRM 在多个基准测试中展现了“以小博大”的惊人实力：在 Sudoku-Extreme（极限数独）测试中，其准确率达到 87.4%；在 Maze-Hard（困难迷宫）中达到 85%；在衡量抽象推理能力的 ARC-AGI 测试中准确率为 45%；ARC-AGI-2 的准确率为 8%。尽管 TRM 使用的参数不到 0.01%，但这些结果仍然超过或接近几种高端大型语言模型的性能，包括 DeepSeek R1 、 Gemini 2.5 Pro 和 o3-mini 。然而，一个重要的前提是，TRM 是专门为解决结构化、可视化的网格类问题（如数独、迷宫和特定解谜任务）而设计的，并非通用的语言聊天模型，它擅长在有明确规则的封闭环境中进行逻辑推理，而非开放式的语言生成。TRM 的成功源于其刻意追求的“少即是多”极简主义设计。研究发现，增加模型层数或大小反而会导致在小数据集上出现过拟合，性能下降。其精简的双层结构与递归深度相结合，实现了最佳效果。TRM 的代码、训练脚本和数据集目前已在 GitHub 上根据 MIT 许可证完全开源，企业和研究人员均可免费使用、修改和部署，甚至用于商业应用。","kind":"news","is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"1","news_tag":"","news_rank":0,"symbols":[],"gpt_button":0,"need_auth":false,"code":"91000000","status":"200"},"commentList":[],"isCommentEnd":true,"newsSizeData":{"likeSize":0,"commentSize":0,"repostSize":0,"favoriteSize":0,"likeStatus":false,"favoriteStatus":false},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":false,"tenantId":"TBCN","deviceId":"web-server-community-laohu8-v3","version":"4.36.2","shortVersion":"4.36.2","platform":"web","vendor":"web","appName":"laohu8","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"href":"/m/news/2574147184","isCrawlerRequest":true}