Meituan has officially released LongCat-2.0, a large-scale MoE language model with 1.6 trillion total parameters and approximately 48 billion activated parameters per token. This model was trained entirely on AI ASIC superpods, spanning millions of accelerator-hours and over 35 trillion tokens, demonstrating capability in frontier-scale training on alternative hardware. LongCat-2.0 introduces LongCat Sparse Attention and was trained on hundreds of billions of tokens with a 1 million token context window, enhancing its performance on coding and agentic tasks. AI
IMPACT This release showcases advanced training capabilities on alternative hardware and a significant increase in context window, potentially influencing future large model development.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
Read on Hugging Face Trending Models →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
