{"rewrite":{"id":"r_98f93331d60af17a82fcd97c","clusterId":"c_a993263aa99b8e1a6d9ee467","slug":"meituan-releases-longcat-2-0-ai-model-trained-entirely-on-domestic-chips","model":"deepseek-v4-flash","headline":"Meituan Releases LongCat-2.0 AI Model Trained Entirely on Domestic Chips","summary":"Chinese company Meituan has released LongCat-2.0, a 1.6-trillion-parameter AI model with a 1-million-token context length, trained and deployed entirely on over 50,000 domestic ASIC superpods. The model achieves performance comparable to Gemini 3.1 Pro and is scheduled for open release on Hugging Face.","whyItMatters":"LongCat-2.0 demonstrates that a large-scale AI model can be built without relying on NVIDIA chips, addressing a key bottleneck for Chinese AI development under US export restrictions.","webCardHtml":"\u003cp\u003eMeituan announced LongCat-2.0, a large MoE language model with 1.6 trillion total parameters and approximately 48 billion activated per token. The model supports a maximum context length of 1 million tokens, matching DeepSeek-V4-Pro\u0026#39;s specifications. Meituan developed LongCat Sparse Attention (LSA), an evolution of DeepSeek Sparse Attention, to accelerate long-context processing without quality loss.\u003c/p\u003e\u003cp\u003eThe entire pipeline from pre-training to inference ran on over 50,000 domestic ASIC superpods. Meituan did not name the chip manufacturer but later confirmed use of Huawei Collective Communication Library (HCCL) for chip-to-chip communication. Because the domestic accelerators have less memory per device than the China-specific NVIDIA H800 (80GB), Meituan addressed the gap through parallelization strategy and memory management.\u003c/p\u003e\u003cp\u003eLongCat-2.0\u0026#39;s performance is comparable to Gemini 3.1 Pro across three evaluation categories: CODE AGENT, GENERAL AGENT, and FOUNDATIONAL. The model is scheduled for open release on Hugging Face, with related code already available on GitHub.\u003c/p\u003e","blueskyPost":"LongCat-2.0 runs on over 50,000 Meituan ASIC superpods, not NVIDIA chips. The model's open release on Hugging Face will let others replicate the domestic-hardware approach.","twitterPost":"LongCat-2.0 matches Gemini 3.1 Pro on domestic ASICs. Open release on Hugging Face invites replication.","threadsPost":"LongCat-2.0 achieves Gemini 3.1 Pro level performance using over 50,000 domestic ASIC superpods, bypassing NVIDIA hardware entirely. Meituan is releasing it openly on Hugging Face, which means other developers can study and reproduce the domestic-chip training pipeline.","newsletterBlurb":"Meituan released LongCat-2.0, a 1.6-trillion-parameter AI model trained and deployed entirely on domestic ASIC superpods, bypassing NVIDIA chips. The model achieves performance comparable to Gemini 3.1 Pro and is scheduled for open release on Hugging Face.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260701-longcat-2-0/\",\"title\":\"Chinese AI Model 'LongCat-2.0' with 1.6 Trillion Parameters Achieves Performance Comparable to Gemini 3.1 Pro Without Using NVIDIA AI Chips\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4307,"outputTokens":664,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1782883735,"createdAt":"2026-07-01T05:21:11.000Z","publishedAt":"2026-07-01T05:24:50.000Z","updatedAt":"2026-07-01T05:24:50.000Z"},"cluster":{"id":"c_a993263aa99b8e1a6d9ee467","canonicalTitle":"NVIDIAのAIチップを使わずにGemini 3.1 Pro並みの性能を実現した1.6兆パラメーターの中国製AIモデル「LongCat-2.0」が登場","representativeArticleId":"a_36b3c4b34755d03f4b7f69ac","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"LongCat-2.0\"],\"studios\":[\"Meituan\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-07-01T04:25:00.000Z","lastSeenAt":"2026-07-01T04:25:00.000Z","updatedAt":"2026-07-01T05:24:51.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260701-longcat-2-0/","title":"NVIDIAのAIチップを使わずにGemini 3.1 Pro並みの性能を実現した1.6兆パラメーターの中国製AIモデル「LongCat-2.0」が登場"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["LongCat-2.0"],"studios":["Meituan"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Meituan released LongCat-2.0, a 1.6-trillion-parameter MoE language model with 48 billion activated parameters per token.","The model supports a 1-million-token context length, matching DeepSeek-V4-Pro.","LongCat-2.0 was trained and deployed entirely on over 50,000 domestic ASIC superpods, using Huawei Collective Communication Library (HCCL) for chip-to-chip communication.","Meituan developed LongCat Sparse Attention (LSA) to accelerate long-context processing without quality loss.","The model's performance is comparable to Gemini 3.1 Pro across CODE AGENT, GENERAL AGENT, and FOUNDATIONAL evaluations, and it is scheduled for open release on Hugging Face."]}