Tag
User benchmarks Qwen3.6-27B on an RTX 3090 using llama.cpp, achieving 35 tok/s generation and 1247 tok/s prompt processing.