The Chinese delivery platform Meituan has released the weights for its LongCat-2.5 model, a multimodal Mixture-of-Experts (MoE) architecture with approximately 1.6 trillion parameters. Notably, this model was trained without relying on Western accelerators, and its deployment guide omits any instructions for NVIDIA GPUs. The model weights are now available under a permissive MIT license. AI
IMPACT Demonstrates alternative training infrastructure for large models, potentially reducing reliance on Western hardware.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →