{"id":362,"date":"2026-07-01T03:05:31","date_gmt":"2026-07-01T03:05:31","guid":{"rendered":"https:\/\/localseobot.ai\/blog\/?p=362"},"modified":"2026-08-02T19:02:40","modified_gmt":"2026-08-02T19:02:40","slug":"meituan-longcat-2-0-trained-on-chinese-chips","status":"publish","type":"post","link":"https:\/\/localseobot.ai\/blog\/meituan-longcat-2-0-trained-on-chinese-chips\/","title":{"rendered":"Meituan&#8217;s LongCat-2.0 Was Trained Entirely on Chinese Chips, the Company Says"},"content":{"rendered":"<p class=\"post-meta-row\"><span class=\"post-meta-time\">6 min read<\/span> \u00b7 <span class=\"post-meta-updated\">Last updated 2026-06-30<\/span><\/p>\n<nav class=\"post-toc\" aria-label=\"Table of contents\"><strong>In this article<\/strong><\/p>\n<ol>\n<li><a href=\"#why-it-matters\">Why It Matters<\/a><\/li>\n<li><a href=\"#whats-new\">What Is LongCat-2.0?<\/a><\/li>\n<li><a href=\"#the-numbers\">What Are the Key Numbers?<\/a><\/li>\n<li><a href=\"#what-comes-next\">What Comes Next<\/a><\/li>\n<li><a href=\"#what-this-means\">What Does This Mean in Practice?<\/a><\/li>\n<li><a href=\"#the-bigger-picture\">What Is the Bigger Picture?<\/a><\/li>\n<\/ol>\n<\/nav>\n<p class=\"wp-block-paragraph\">Meituan launched LongCat-2.0 this week, a 1.6-trillion-parameter AI model the company says is the first of its scale trained end-to-end on domestically developed Chinese chips. The claim is aimed directly at US export controls that have restricted Nvidia&#8217;s most advanced hardware from reaching China, and it positions a food-delivery and services giant as an unlikely flag-bearer for Chinese frontier AI.<\/p>\n<h2 class=\"wp-block-heading\" id=\"why-it-matters\">Why It Matters<\/h2>\n<p class=\"wp-block-paragraph\">For years the open question over China&#8217;s AI sector has been whether it can build frontier-scale models without Nvidia. Washington restricts exports of the most advanced chips on national security grounds, betting that limited access to such chips would slow China&#8217;s progress. A 1.6-trillion-parameter model that Meituan says was both trained and served on home-grown hardware is a direct test of that bet. If the claim holds, the single biggest lever the US has used to contain Chinese AI looks less decisive than it did.<\/p>\n<h2 class=\"wp-block-heading\" id=\"whats-new\">What Is LongCat-2.0?<\/h2>\n<p class=\"wp-block-paragraph\">LongCat-2.0 carries 1.6 trillion parameters and a context window of one million tokens, and Meituan says its performance is comparable to Google&#8217;s Gemini 3.1 Pro, released in February. The company describes it as &#8220;the industry&#8217;s first trillion-parameter model to complete end-to-end training and inference on a 50,000-chip domestic compute cluster.&#8221; The model has been open-sourced, putting the weights in the hands of anyone who wants to run or scrutinise them.<\/p>\n<p class=\"wp-block-paragraph\">The crucial phrase is &#8220;end-to-end.&#8221; Plenty of Chinese models already run inference, the comparatively light task of answering a query once a model is built, on domestic hardware. Pre-training is the heavy part, the computationally brutal process in which a model digests vast data sets to learn its basic patterns, and it is where the most advanced chips have mattered most. Meituan&#8217;s claim that LongCat-2.0 was both pre-trained and served on domestic silicon is what makes the announcement more than a marketing line.<\/p>\n<h2 class=\"wp-block-heading\" id=\"the-numbers\">What Are the Key Numbers?<\/h2>\n<ul class=\"wp-block-list\">\n<li><strong>1.6 trillion parameters,<\/strong> putting LongCat-2.0 among the largest models publicly announced.<\/li>\n<li><strong>1 million token context window,<\/strong> for long-document and long-session work.<\/li>\n<li><strong>50,000-chip domestic compute cluster,<\/strong> used for what Meituan calls end-to-end training and inference.<\/li>\n<li><strong>Comparable to Google Gemini 3.1 Pro,<\/strong> by Meituan&#8217;s own account, on the benchmarks it cites.<\/li>\n<li><strong>Fully open-sourced weights,<\/strong> available for anyone to run or scrutinise.<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>&#8220;The industry&#8217;s first trillion-parameter model to complete end-to-end training and inference on a 50,000-chip domestic compute cluster.&#8221;<\/p>\n<footer>Meituan, describing LongCat-2.0<\/footer>\n<\/blockquote>\n<h2 class=\"wp-block-heading\" id=\"what-comes-next\">What Comes Next<\/h2>\n<p class=\"wp-block-paragraph\">Independent verification will come from the open-source community, which can now run LongCat-2.0 against the benchmarks Meituan cites and test whether it genuinely matches a model like Gemini 3.1 Pro. The training-hardware claim is harder for outsiders to confirm, since it rests on Meituan&#8217;s account of its own infrastructure, and that caveat is worth holding in mind alongside the company&#8217;s confidence. LongCat-2.0 is the software counterpart to a broader hardware push: China recently claimed the supercomputing crown without US chips, and domestic challengers such as Alibaba&#8217;s T-Head unit are promoting home-grown accelerators like the Zhenwu M890 GPU.<\/p>\n<figure class=\"wp-block-pullquote\">\n<blockquote class=\"pull-quote\">\n<p>Each frontier-scale model trained without American hardware narrows the gap the export controls were meant to widen.<\/p>\n<\/blockquote>\n<\/figure>\n<h2 class=\"wp-block-heading\" id=\"what-this-means\">What Does This Mean in Practice?<\/h2>\n<p class=\"wp-block-paragraph\">For anyone building on AI, the story is a reminder that the supply of capable models is globalising, not narrowing. An open-source model at this scale lowers the cost of frontier capability and widens the pool of providers beyond the familiar US names. Meituan itself is an unlikely flag-bearer, better known for food delivery than frontier AI, and its motive is concrete: routing, demand forecasting, and customer service all run on compute, and a model trained on domestic silicon insulates that compute from the next turn of the export-control screw. The practical takeaway for teams elsewhere is to keep an eye on open-weight models from outside the US, because the best price-to-performance option may increasingly come from an unexpected source.<\/p>\n<h2 class=\"wp-block-heading\" id=\"the-bigger-picture\">What Is the Bigger Picture?<\/h2>\n<p class=\"wp-block-paragraph\">At its base, the AI contest between China and the United States has become a race over chips. Export controls were designed to widen America&#8217;s lead by denying China the hardware to train the largest models. Every credible claim of a frontier-scale model trained on domestic silicon chips away at that strategy. Meituan&#8217;s announcement is one more data point in a contest Washington built its restrictions to win, and that Beijing is determined to prove it can run on its own terms.<\/p>\n<h2 id=\"faq\">FAQ<\/h2>\n<h3>What is LongCat-2.0?<\/h3>\n<p>LongCat-2.0 is Meituan&#8217;s new large language model, a 1.6-trillion-parameter system with a one-million-token context window. It has been open-sourced, and Meituan says its performance is comparable to Google&#8217;s Gemini 3.1 Pro.<\/p>\n<h3>Why is training on domestic Chinese chips significant?<\/h3>\n<p>Pre-training is the most compute-intensive stage of building a model and the point where the most advanced chips have mattered most. Completing it end-to-end on domestically developed hardware suggests China can build frontier-scale models without US silicon, the exact outcome that export controls were meant to prevent.<\/p>\n<h3>Has the training-on-domestic-chips claim been independently verified?<\/h3>\n<p>Not yet. The open-source community can test the benchmark claims now that the weights are public, but the training-hardware claim rests on Meituan&#8217;s own account of its infrastructure and is harder for outsiders to confirm directly.<\/p>\n<p><script type=\"application\/ld+json\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"headline\":\"Meituan's LongCat-2.0 Was Trained Entirely on Chinese Chips, the Company Says\",\"description\":\"Meituan says its 1.6-trillion-parameter LongCat-2.0 is the first model of its scale trained end-to-end on a 50,000-chip domestic cluster, testing US export controls.\",\"datePublished\":\"2026-07-19T06:47:02.273Z\",\"publisher\":{\"@type\":\"Organization\",\"name\":\"LocalSEOBot\"}},{\"@type\":\"FAQPage\",\"mainEntity\":[{\"@type\":\"Question\",\"name\":\"What is LongCat-2.0?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"LongCat-2.0 is Meituan's new large language model, a 1.6-trillion-parameter system with a one-million-token context window. It has been open-sourced, and Meituan says its performance is comparable to Google's Gemini 3.1 Pro.\"}},{\"@type\":\"Question\",\"name\":\"Why is training on domestic Chinese chips significant?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Pre-training is the most compute-intensive stage of building a model and the point where the most advanced chips have mattered most. Completing it end-to-end on domestically developed hardware suggests China can build frontier-scale models without US silicon, the exact outcome that export controls were meant to prevent.\"}},{\"@type\":\"Question\",\"name\":\"Has the training-on-domestic-chips claim been independently verified?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Not yet. The open-source community can test the benchmark claims now that the weights are public, but the training-hardware claim rests on Meituan's own account of its infrastructure and is harder for outsiders to confirm directly.\"}}]}]}<\/script><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Meituan says its 1.6-trillion-parameter LongCat-2.0 is the first model of its scale trained end-to-end on a 50,000-chip domestic cluster, testing US export controls.<\/p>\n","protected":false},"author":2,"featured_media":363,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-362","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news"],"_links":{"self":[{"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/posts\/362","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/comments?post=362"}],"version-history":[{"count":3,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/posts\/362\/revisions"}],"predecessor-version":[{"id":725,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/posts\/362\/revisions\/725"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/media\/363"}],"wp:attachment":[{"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/media?parent=362"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/categories?post=362"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localseobot.ai\/blog\/wp-json\/wp\/v2\/tags?post=362"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}