Z.ai's Open-Weight GLM-5.2 Posts Frontier-Adjacent Coding Scores as Tang Rebuts Musk's 'One Year' China Timeline

Tom's Hardware·high signal

Independent benchmarks for the MIT-licensed GLM-5.2 (744B params / 40B active MoE, 1M-token context) landed this week: 81.0 on Terminal-Bench 2.1 (within 4 points of Claude Opus 4.8's 85.0) and 62.1 on SWE-bench Pro, ahead of GPT-5.5's 58.6. On June 19–20, Z.ai co-founder Jie Tang publicly rebutted Elon Musk's prediction that China is a year from a Fable-5-class model, saying 'it won't take that long.' For builders: a permissively-licensed, self-hostable model with day-one support for Claude Code, Cline, and Roo Code is now the strongest Claude alternative for coding agents.