Z.ai, a Chinese AI developer, launched ZCode, an agentic coding tool built on its GLM-5.2 model that plans, writes, and tests code with little human input, taking on GitHub Copilot, Cursor, and Anthropic’s Claude Code. GLM-5.2, the open-weight model beneath the tool, is a mixture-of-experts system with 744 billion parameters that Z.ai released in June under an MIT license, priced at $1.40 per million input tokens and $4.40 per million output, far under what Anthropic charges for comparable work. Benchmarks cited by Z.ai put GLM-5.2 second on Code Arena behind Anthropic’s Claude Fable 5, one point behind Claude Opus 4.8 on FrontierSWE, and ahead of OpenAI’s GPT-5.5. Z.ai built the model on Huawei silicon instead of American chips, part of China’s drive for AI infrastructure free of U.S. technology as it courts enterprise buyers abroad.
