Tag
A deliberately empty 16.5-trillion-parameter model uploaded to Hugging Face exposes that parameter counts are computed from safetensors headers alone, and that Xet's content-defined deduplication reduces upload bandwidth enormously while storage quota still bills the full logical size.
An opinion arguing that Anthropic and OpenAI's advantage is scale rather than secret sauce, with open models like DeepSeek V4 and Kimi K3 catching up as parameter sizes increase.
LongCat-2.0 is a large-scale Mixture-of-Experts (MoE) model with 1.6 trillion total parameters and 48 billion active parameters.