{"id":38131,"date":"2025-06-23T21:19:56","date_gmt":"2025-06-23T13:19:56","guid":{"rendered":"https:\/\/www.1ai.net\/?p=38131"},"modified":"2025-06-23T21:19:56","modified_gmt":"2025-06-23T13:19:56","slug":"%e5%9b%bd%e5%86%85%e9%a6%96%e4%b8%aa%ef%bc%9a%e7%bd%91%e6%98%93%e6%9c%89%e9%81%93%e5%bc%80%e6%ba%90%e5%ad%90%e6%9b%b0-3-%e6%95%b0%e5%ad%a6%e6%a8%a1%e5%9e%8b%ef%bc%8c%e5%8d%95","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/38131.html","title":{"rendered":"First in China: NetEase has open-sourced \"Ziyi 3 Math Model\", which can run on a single consumer GPU"},"content":{"rendered":"<p>June 23rd.<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e7%bd%91%e6%98%93%e6%9c%89%e9%81%93\" title=\"[Sees articles with [accessible] labels]\" target=\"_blank\" >NetEase Youtube<\/a>Announced today<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%bc%80%e6%ba%90\" title=\"[View articles tagged with [open source]]\" target=\"_blank\" >Open Source<\/a>The \"Ziyi 3\" series of large models.<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e6%95%b0%e5%ad%a6%e6%a8%a1%e5%9e%8b\" title=\"_Other Organiser\" target=\"_blank\" >Mathematical model<\/a>(Confucius3-Math), claiming to be the<strong>China's first open-source reasoning model focused on math education that can run efficiently on a single consumer GPU<\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-38132\" title=\"a3146cc8j00syb8zs005hd000u000hep\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/a3146cc8j00syb8zs005hd000u000hep.jpg\" alt=\"a3146cc8j00syb8zs005hd000u000hep\" width=\"1080\" height=\"626\" \/><\/p>\n<p><a href=\"https:\/\/www.1ai.net\/en\/tag\/%e7%bd%91%e6%98%93\" title=\"[Sees articles with [Wepwire] labels]\" target=\"_blank\" >NetEase<\/a>The official test data from Yodo shows that on the CK12-math (Internal), GAOKAO-Bench (Math), MathBench (K12), and MATH500 datasets, the<strong>14B The lightweight \"Say 3 Mathematical Model\" scores higher than general-purpose big models like DeepSeek-R1 in every category.<\/strong>In particular, it scored 98.5 on the GAOKAO-Bench (Math), a large model assessment framework based on high school math questions. Especially in the GAOKAO-Bench (Math), which is a large model assessment framework based on high school math questions, the \"Ziyi 3 Math Model\" scored 98.5 points.<\/p>\n<p>According to NetEase, the ZiYi 3 Mathematical Model achieves better cost-effectiveness through large-scale augmented learning and a series of innovative algorithms -- the most cost-effective way to improve the efficiency of the system.\u00a0<strong>It costs only $26,000 to train (note: current exchange rate is about RMB 187,000) and has about 15 times the inference performance of DeepSeek R1<\/strong>.<\/p>\n<p>In addition, the fact that it runs on consumer GPUs (such as the RTX 4090D) lowers the barriers to AI applications in education scenarios, with a cost of service as low as $0.15 per million tokens, much lower than most general-purpose macromodels today. This means that a wider range of educational institutions and developers can deploy professional-grade math AI applications at low cost.<\/p>\n<p>1AI attached open source address below:<\/p>\n<p>https:\/\/github.com\/netease-youdao\/Confucius3-Math<\/p>\n<p>Demo Address:<\/p>\n<p>https:\/\/confucius.youdao.com\/<\/p>\n<p>Paper address:<\/p>\n<p>https:\/\/github.com\/netease-youdao\/Confucius3-Math\/blob\/main\/Confucius3-Math.pdf<\/p>","protected":false},"excerpt":{"rendered":"<p>On June 23rd, the Internet announced today a mathematical model (Confucius 3-Math) of the large series of open source \u201cSubsistence 3\u201d models, claiming to be the first open source reasoning model in the country to focus on mathematics education and to operate efficiently on single-scale consumption-grade GPUs. Data from online easy-to-access official tests show that the 14B lightweight \"Subsistence 3 Mathematical Models\" scores are higher than those of large generic models such as DeepSeek-R1 in data sets such as CK12-math, GAOKAO-Bench (Math), MathBench (K12) and MATH500. Especially the big model of GAOKAO-Bench (Math) based on advanced mathematics<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[219,4087,1276,5613],"collection":[],"class_list":["post-38131","post","type-post","status-publish","format-standard","hentry","category-news","tag-219","tag-4087","tag-1276","tag-5613"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/38131","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=38131"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/38131\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=38131"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=38131"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=38131"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=38131"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}