{"id":39294,"date":"2025-07-12T11:29:05","date_gmt":"2025-07-12T03:29:05","guid":{"rendered":"https:\/\/www.1ai.net\/?p=39294"},"modified":"2025-07-12T11:29:05","modified_gmt":"2025-07-12T03:29:05","slug":"%e6%9c%88%e4%b9%8b%e6%9a%97%e9%9d%a2-kimi-k2-%e5%8f%91%e5%b8%83%e5%b9%b6%e5%bc%80%e6%ba%90%ef%bc%8c%e6%93%85%e9%95%bf%e4%bb%a3%e7%a0%81%e4%b8%8e-agentic-%e4%bb%bb%e5%8a%a1","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/39294.html","title":{"rendered":"Dark Side of the Moon Kimi K2 released and open source, specializes in code and Agentic tasks"},"content":{"rendered":"<p>July 12 News.<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e6%9c%88%e4%b9%8b%e6%9a%97%e9%9d%a2\" title=\"[Sees articles with labels]\" target=\"_blank\" >Dark Side of the Moon<\/a>It was officially released last night.\u00a0<strong><a href=\"https:\/\/www.1ai.net\/en\/tag\/kimi\" title=\"[View articles tagged with [Kimi]]\" target=\"_blank\" >Kimi<\/a> K2 model and synchronize<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><\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-39295\" title=\"c67466aaj00sz9ob4004pd000u000crp\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/07\/c67466aaj00sz9ob4004pd000u000crp.jpg\" alt=\"c67466aaj00sz9ob4004pd000u000crp\" width=\"1080\" height=\"459\" \/><\/p>\n<p><strong>Kimi K2 is a MoE architectural base model with stronger code capabilities and better generalized Agent tasks<\/strong>The total parameters are 1T and the activation parameters are 32B.<\/p>\n<p>In benchmark performance tests such as SWE Bench Verified, Tau2, and AceBench, Kimi K2 has achieved SOTA scores among open source models, demonstrating leading capabilities in code, Agent, and mathematical reasoning tasks.<\/p>\n<p>The pre-training phase of Kimi K2 uses the MuonClip optimizer to achieve stable and efficient training of trillion-parameter models, which effectively improves the efficiency of Token utilization and finds new Scaling space in the context of high-quality human data becoming a bottleneck.<\/p>\n<p>Kimi K2 has achieved excellent performance in benchmark performance tests in the three competency dimensions of Agentic Coding, Tool Use, and Math &amp; Reasoning.<\/p>\n<p>In addition to benchmark performance tests, Kimi K2 also demonstrates greater capability generalization and utility in several real-world scenarios.<\/p>\n<p>From now on, you can visit the official website kimi.com or download the Kimi App to experience the new Kimi K2 models. The API service has also been launched simultaneously, providing a Chat API interface compatible with OpenAI and Anthropic, which allows users to switch their commonly used big model tools to Kimi K2.<\/p>\n<p>Kimi K2's API service is now fully online and supports<strong>Maximum 128K contexts<\/strong>, with greater versatility and tool-calling capabilities. The billing program is as follows:<\/p>\n<ul>\n<li>Per million input tokens: $4<\/li>\n<li>Per million output tokens: $16<\/li>\n<\/ul>\n<p>Dark Side of the Moon has synchronized and open-sourced two model versions from the Kimi K2 series:<\/p>\n<ul>\n<li><strong>Kimi-K2-Base<\/strong>: Basic pre-trained model not fine-tuned by instructions, suitable for scientific research and customized scenarios;<\/li>\n<li><strong>Kimi-K2-Instruct<\/strong>: A fine-tuned version of the General Instruction (non-thinking model) that performs well in most Q&amp;A and Agent tasks.<\/li>\n<\/ul>\n<p>1AI Attached open source address:<\/p>\n<p>https:\/\/huggingface.co\/moonshotai\/Kimi-K2-Instruct<\/p>","protected":false},"excerpt":{"rendered":"<p>July 12, 2012 - Dark Side of the Moon officially released the Kimi K2 model last night and open-sourced it simultaneously. Kimi K2 is a MoE architecture base model with stronger code capabilities and better at generalized Agent tasks, with 1T total parameters and 32B activation parameters. In benchmark performance tests such as SWE Bench Verified, Tau2, AceBench, etc., Kimi K2 has achieved SOTA scores among the open-source models, demonstrating its leading capabilities in code, Agent, mathematical reasoning tasks. In the pre-training phase of Kimi K2, the MuonClip optimizer is used to achieve stable and efficient training of trillion-parameter models, and the high quality human data is the bottleneck.<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[1814,219,1168],"collection":[],"class_list":["post-39294","post","type-post","status-publish","format-standard","hentry","category-news","tag-kimi","tag-219","tag-1168"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/39294","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=39294"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/39294\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=39294"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=39294"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=39294"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=39294"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}