CHINESE OPEN-SOURCE MODEL DETHRONES CLAUDE AND GPT — GLM-5.1 RUNS 8 HOURS STRAIGHT ON CODING TASKS Z.ai’s GLM-5.1 just crashed the AI benchmark party, outperforming Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro — and crucially, it’s fully open source with 754 billion parameters. The model ran for 8 straight hours completing over 6,000 tool calls on a single complex coding task, which is either impressive or deeply unsettling depending on your job description. Silicon Valley’s closed-source labs, which charge a fortune for access, just received another very public memo from a Chinese startup now worth $52 billion on the Hong Kong Stock Exchange. Keywords: GLM-5.1, open source AI, Z.ai, SWE-Bench benchmark, AI coding model