Meituan's open-weights 1.6T-parameter mixture-of-experts LLM (48B active), aimed at agentic coding and reasoning tasks.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
LongCat 2.0 is a large language model from Meituan, part of its LongCat model family. According to a YouTube creator summary, it is a mixture-of-experts model with about 1.6 trillion total parameters and 48 billion active per token, released under the MIT license. The same source says it was trained entirely on AI ASIC super pods rather than Nvidia GPUs.
Reported strengths are agentic coding benchmarks, where it is described as close to leading closed models. These claims come from a third-party video and were not verified against official benchmark pages. The official site (longcat.chat) is a chat interface; the scraped page text gave no technical details.
Because of its size, the weights are reported to exceed 2 TB even in FP8, so self-hosting needs multi-node server hardware. Most users will probably reach it through the hosted chat or an API rather than running it locally.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source 1.6T parameter MoE model (48B active) from Meituan, trained entirely on AI ASIC super pods without Nvidia GPUs. It is near top closed models on agentic coding benchmarks, MIT licensed, but over 2TB even in FP8.
Closest neighbours by meaning, not just by category.
Model · Meituan
Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.
Model
35B Mixture-of-Experts open-weights model from InternScience for long-horizon agentic search, engineering and tool use.
Model
Open-weights 35B-A3B agentic model post-trained from Qwen3.6-35B-A3B for search, coding and scientific research.
Model · Alibaba
Alibaba's open-weights 2.4T-parameter MoE language model (95B active) from the Qwen3.8 family, with a hosted Max version.
Model · MiniMax
Open-weights native multimodal MoE model from MiniMax, ~428B parameters (~23B active) with 1M-token context.
Model
Open 7.39B dense model for math reasoning and agentic search, with a 256K context and the full training pipeline released.
