Meituan Releases LongCat 2.0, a 1.6 Trillion Parameter Open-Source AI Model Trained on Huawei Chips
Meituan released LongCat 2.0, a 1.6 trillion parameter open-source large language model licensed under the MIT agreement. The mixture-of-experts architecture uses 48 billion active parameters and supports a 1 million token context window.
The model outperforms Google’s Gemini 3.1 Pro on several coding benchmarks and was trained on AI ASIC superpods widely believed to be Huawei Ascend 910C processors. Weights and inference code became available following a two-month run on OpenRouter under a different identifier, prompting discussion over what the hardware shift means for Nvidia’s dominance in frontier model training.
From the sources (13 posts)
@eliebakouchlongcat 2.0 (1.6T, ~48B active) weights are now open under MIT license interesting design choice compared to previous iteration is that they keep the same attention shape and REDUCE the number of zero communication experts from 256 to 128
@multimodalartLe Chaton Long is here! The 1.8T behemoth of Meituan's LongCat-2.0 model is now open weights. A 1T+ parameter model trained entirely on chinese ASIC
@victormustarRT @Meituan_LongCat: 🐱 LongCat-2.0 is now fully open-source — MIT licensed, no restrictions. Since our launch a few days ago, the response…
@stevibeRT @Meituan_LongCat: 🐱 LongCat-2.0 is now fully open-source — MIT licensed, no restrictions. Since our launch a few days ago, the response…
@jordannanosthe npu inference repo in here confirms that they are using Ascend 910B
@lmsysorgRT @sgl_project: Owl Alpha, the model that hit Top 3 Call Volume on OpenRouter right after launch, is now open source. It's LongCat 2.0 by…
@sgl_projectOwl Alpha, the model that hit Top 3 Call Volume on OpenRouter right after launch, is now open source. It's LongCat 2.0 by @Meituan_LongCat, exclusively supported by SGLang. Download the weights and deploy it yourself. Everything you need
@thezachmuellerRT @Meituan_LongCat: 🐱 LongCat-2.0 is now fully open-source — MIT licensed, no restrictions. Since our launch a few days ago, the response…
@poezhao0605Meituan just open-sourced a 1.6 trillion parameter model for AI coding agents. It is the first trillion-parameter model to complete inference on a 50,000 domestic chip cluster. BF16, FP8, and INT8 versions ship together.
@meituan_longcat🐱 LongCat-2.0 is now fully open-source — MIT licensed, no restrictions. Since our launch a few days ago, the response from the community has been incredible. Thank you for all the feedback, discussions, and interest. Today, we’re releasin
@mtsliveRT @gbrl_dick: last week, Meituan officially released LongCat 2.0, an open-source model that performs better than Gemini 3.1 Pro on a numbe…
@liangsaysRT @gbrl_dick: last week, Meituan officially released LongCat 2.0, an open-source model that performs better than Gemini 3.1 Pro on a numbe…
@gbrl_dicklast week, Meituan officially released LongCat 2.0, an open-source model that performs better than Gemini 3.1 Pro on a number of coding benchmarks. but there was two surprises—first, the model, which performs similarly to DeepSeek V4, had