{"id":25310,"date":"2025-07-14T17:12:17","date_gmt":"2025-07-14T14:12:17","guid":{"rendered":"https:\/\/forklog.com\/en\/chinese-ai-startup-moonshot-ai-launches-new-open-model\/"},"modified":"2025-07-14T17:12:17","modified_gmt":"2025-07-14T14:12:17","slug":"chinese-ai-startup-moonshot-ai-launches-new-open-model","status":"publish","type":"post","link":"https:\/\/u1f987.com\/en\/chinese-ai-startup-moonshot-ai-launches-new-open-model\/","title":{"rendered":"Chinese AI Startup Moonshot AI Launches New Open Model"},"content":{"rendered":"<p>Chinese AI startup Moonshot AI, backed by Alibaba, has unveiled a new open-source model, Kimi K2. It is claimed to excel in advanced knowledge, mathematics, programming, and agentic tasks.<\/p>\n<blockquote class=\"twitter-tweet\">\n<p lang=\"en\" dir=\"ltr\">\ud83d\ude80 Hello, Kimi K2! Open-Source Agentic Model!<br \/>\ud83d\udd39 1T total \/ 32B active MoE model<br \/>\ud83d\udd39 SOTA on SWE Bench Verified, Tau2 &#038; AceBench among open models<br \/>\ud83d\udd39Strong in coding and agentic tasks<br \/>\ud83d\udc24 Multimodal &#038; thought-mode not supported for now<\/p>\n<p>With Kimi K2, advanced agentic intelligence\u2026 <a href=\"https:\/\/t.co\/PlRQNrg9JL\">pic.twitter.com\/PlRQNrg9JL<\/a><\/p>\n<p>\u2014 Kimi.ai (@Kimi_Moonshot) <a href=\"https:\/\/twitter.com\/Kimi_Moonshot\/status\/1943687594560332025?ref_src=twsrc%5Etfw\">July 11, 2025<\/a><\/p><\/blockquote>\n<p> <script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<p>In its announcement, the startup stated that its neural network surpassed Claude Opus 4 in two benchmarks and outperformed OpenAI&#8217;s GPT-4.1 in several metrics. Additionally, the model offers lower token processing costs:<\/p>\n<ul class=\"wp-block-list\">\n<li>$0.15 per 1 million input tokens;<\/li>\n<li>$2.5 per 1 million output tokens.<\/li>\n<\/ul>\n<p>In comparison, Claude Opus 4 charges $15 per 1 million input tokens and $75 per 1 million output tokens\u2014100 and 30 times more, respectively. For OpenAI&#8217;s GPT-4.1, the figures are $2 and $8, respectively.<\/p>\n<p>Kimi K2 is developed based on the <span data-descr=\"an approach in machine learning where an AI model is divided into separate submodules (experts), each processing its own part of the input data. This significantly reduces computational costs during pre-training and speeds up the model's operation during use.\" class=\"old_tooltip\">Mixture-of-Experts<\/span> architecture and includes 1 trillion parameters, of which 32 billion are active\u2014representing specialized computational blocks used for specific tasks.<\/p>\n<p>Kimi K2 is available for free through the Kimi app and web interface. The company has released the source code for two versions:<\/p>\n<ul class=\"wp-block-list\">\n<li>Kimi-K2-Base\u2014a basic model intended for researchers and developers who need full control over customization and tuning;<\/li>\n<li>Kimi-K2-Instruct\u2014a post-trained version optimized for general use, including chatbots and agentic AI scenarios.<\/li>\n<\/ul>\n<p>Initial feedback on Kimi K2 in English and Chinese social media is mostly positive, although some point out the common issue of hallucinations in the AI segment.<\/p>\n<blockquote class=\"twitter-tweet\">\n<p lang=\"en\" dir=\"ltr\">Kimi K2 is so good at tool calling and agentic loops, can call multiple tools in parallel and reliably, and knows &#8220;when to stop&#8221;, which is another important property. <\/p>\n<p>It&#8217;s the first model I feel comfortable using in production since Claude 3.5 Sonnet. <a href=\"https:\/\/t.co\/TcEkPlBMuk\">pic.twitter.com\/TcEkPlBMuk<\/a><\/p>\n<p>\u2014 Pietro Schirano (@skirano) <a href=\"https:\/\/twitter.com\/skirano\/status\/1944475540951621890?ref_src=twsrc%5Etfw\">July 13, 2025<\/a><\/p><\/blockquote>\n<p> <script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>&#8220;It&#8217;s the first model I feel comfortable using in production since Claude 3.5 Sonnet,&#8221; wrote MagicPath startup founder Pietro Schirano on X.<\/p>\n<\/blockquote>\n<h2 class=\"wp-block-heading\"><strong>Open AI\u2014A Growing Trend<\/strong><\/h2>\n<p>The new model reflects a broader industry trend towards open-source code, enabling both startups and major tech companies to enhance efficiency and accelerate the deployment of their AI products. This approach involves providing public access to a program&#8217;s source code, allowing third-party developers to:<\/p>\n<ul class=\"wp-block-list\">\n<li>modify and refine the model&#8217;s architecture;<\/li>\n<li>fix bugs;<\/li>\n<li>scale functionality;<\/li>\n<li>implement their own improvements and adaptations for specific tasks.<\/li>\n<\/ul>\n<p>Among notable Chinese startups, open-source code is <a href=\"https:\/\/u1f987.com\/en\/news\/deepseek-unveils-mathematical-ai-model-prover-v2\">offered<\/a> by DeepSeek and Tencent.<\/p>\n<p><script async src=\"https:\/\/telegram.org\/js\/telegram-widget.js?22\" data-telegram-post=\"forklogAI\/5761\" data-width=\"100%\"><\/script><\/p>\n<p>As for American competitors, <a href=\"https:\/\/u1f987.com\/en\/news\/meta-unveils-new-series-of-ai-models-llama-4\">Meta<\/a> and partially <a href=\"https:\/\/u1f987.com\/en\/news\/google-introduces-new-open-ai-models\">Google<\/a> have taken this path.<\/p>\n<p>Meanwhile, OpenAI CEO Sam Altman announced the postponement of the launch of their own open AI model indefinitely, citing the need for additional safety tests.<\/p>\n<blockquote class=\"twitter-tweet\">\n<p lang=\"en\" dir=\"ltr\">we planned to launch our open-weight model next week.<\/p>\n<p>we are delaying it; we need time to run additional safety tests and review high-risk areas. we are not yet sure how long it will take us.<\/p>\n<p>while we trust the community will build great things with this model, once weights are\u2026<\/p>\n<p>\u2014 Sam Altman (@sama) <a href=\"https:\/\/twitter.com\/sama\/status\/1943837550369812814?ref_src=twsrc%5Etfw\">July 12, 2025<\/a><\/p><\/blockquote>\n<p> <script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<p>Back in March 2024, Elon Musk <a href=\"https:\/\/u1f987.com\/en\/news\/elon-musk-announces-open-sourcing-of-groks-code\">announced<\/a> the release of Grok&#8217;s source code.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Chinese AI startup Moonshot AI, backed by Alibaba, has unveiled a new open-source model, Kimi K2. It is claimed to excel in advanced knowledge, mathematics, programming, and agentic tasks. \ud83d\ude80 Hello, Kimi K2! Open-Source Agentic Model!\ud83d\udd39 1T total \/ 32B active MoE model\ud83d\udd39 SOTA on SWE Bench Verified, Tau2 &#038; AceBench among open models\ud83d\udd39Strong in [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":25309,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"select":"","news_style_id":"","cryptorium_level":"","_short_excerpt_text":"","creation_source":"","_metatest_mainpost_news_update":false,"footnotes":""},"categories":[3],"tags":[438,133],"class_list":["post-25310","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-and-analysis","tag-artificial-intelligence","tag-china"],"aioseo_notices":[],"amp_enabled":true,"views":"62","promo_type":"","layout_type":"","short_excerpt":"","is_update":"","_links":{"self":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts\/25310","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/comments?post=25310"}],"version-history":[{"count":0,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts\/25310\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/media\/25309"}],"wp:attachment":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/media?parent=25310"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/categories?post=25310"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/tags?post=25310"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}