{"id":49472,"date":"2026-02-02T13:21:31","date_gmt":"2026-02-02T05:21:31","guid":{"rendered":"https:\/\/www.1ai.net\/?p=49472"},"modified":"2026-02-02T13:21:31","modified_gmt":"2026-02-02T05:21:31","slug":"%e4%b8%ba-agent-%e8%80%8c%e7%94%9f%ef%bc%8c%e9%98%b6%e8%b7%83%e6%98%9f%e8%be%b0%e5%8f%91%e5%b8%83%e5%bc%80%e6%ba%90%e5%9f%ba%e5%ba%a7%e6%a8%a1%e5%9e%8b-step-3-5-flash","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/49472.html","title":{"rendered":"Born for Agent, Step 3.5 Flash releases the open source base model"},"content":{"rendered":"<p>February 2 News.<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e9%98%b6%e8%b7%83%e6%98%9f%e8%be%b0\" title=\"[View articles tagged with [Step Star]]\" target=\"_blank\" >Step Star<\/a>The latest, most powerful open source under the flag was released<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%9f%ba%e5%ba%a7%e6%a8%a1%e5%9e%8b\" title=\"[Sees articles with labels]\" target=\"_blank\" >Base base model<\/a> Step 3.5 Flash, officially known as having strong reasoning skills and <a href=\"https:\/\/www.1ai.net\/en\/tag\/agent\" title=\"[View articles tagged with [Agent]]\" target=\"_blank\" >Agent<\/a> Smart, born for Agent\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-49473\" title=\"131ceb77j00t9tg755002gd000u000ep\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/02\/131ceb77j00t9tg75002gd000u000ezp.jpg\" alt=\"131ceb77j00t9tg755002gd000u000ep\" width=\"1080\" height=\"539\" \/><\/p>\n<p>performance vs model size<\/p>\n<p>1AI HAS THE FOLLOWING OFFICIAL DESCRIPTION:<\/p>\n<blockquote>\n<ul>\n<li>The core highlights are as follows:<\/li>\n<\/ul>\n<\/blockquote>\n<ul>\n<li>FASTER: HIGHEST REASONING 350 TPS<\/li>\n<li>Stronger: Compared closed-source models to Agent scenes and mathematical missions<\/li>\n<li>Stabler: qualified for complex, long chain tasks<\/li>\n<\/ul>\n<blockquote>\n<ul>\n<li>We need not only \u201csmarter\u201d models, but also \u201ctrustworthy\u201d, \u201cfast-response\u201d, and \u201ccost-controllable\u201d models. Step 3.5 Flash is here for this reason. We use:<\/li>\n<\/ul>\n<\/blockquote>\n<ul>\n<li>Rare MoE structure: each token only activates about 11 billion parameters (total 1960 billion parameters)<\/li>\n<li>MTP-3: Model one-time projection 3 Token, double efficiency<\/li>\n<li>3:1: Slide Window with Global Attention (SWA + Full Attention): Focus only in long text, significantly lower calculation costs and efficiently handle 256K context<\/li>\n<\/ul>\n<p>Step 3.5 Flash is now fully online, and the official report also states that the Step 4 model training has been initiated and that you are invited to participate in the creation of a joint venture\u3002<\/p>","protected":false},"excerpt":{"rendered":"<p>On February 2nd, Step 3.5 Flash, the latest and most powerful base model under the flag, was released by the step-to-step star, officially known as having a strong reasoning capability and Agent intelligence, born for Agent. The performance of the vs model size 1AI with the official introduction is as follows: The core highlights are as follows: Faster: the highest rate of reasoning 350 TPS (single-request code type jobs) is stronger: the close-source model is more stable in the Agent scene and mathematical missions: we need not only more \"smart\" models, but more \"trustworthy\", \"fast-response\" and \"cost-controlable\" models. Step 3.5 Flash is the reason why we use the thin MoE architecture<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[1405,8217,1893],"collection":[],"class_list":["post-49472","post","type-post","status-publish","format-standard","hentry","category-news","tag-agent","tag-8217","tag-1893"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/49472","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=49472"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/49472\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=49472"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=49472"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=49472"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=49472"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}