Meituan has developed LongCat-2.0, a 1.6 trillion parameter language model trained entirely on domestic Chinese AI chips, avoiding NVIDIA hardware. The model's efficiency stems from its Mixture-of-Experts architecture, which utilizes only 3% of its parameters per token. This design significantly reduces the per-step workload and network traffic, enabling training on less powerful, non-NVIDIA accelerators. AI
IMPACT Demonstrates feasibility of training large models without top-tier accelerators by optimizing parameter utilization.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
