Summary
On July 2, 2026, Beijing-based Z.ai launched ZCode, a desktop agentic coding environment that serves as the official harness for its open-weight GLM-5.2 model. GLM-5.2 (MIT-licensed, a 744B-parameter mixture-of-experts with ~40B active per token and a 1M-token context) scores 62.1 on SWE-bench Pro, ahead of GPT-5.5 (58.6) and within four points of Claude Opus 4.8 (66.0). In its first full week, GLM-5.2 saw roughly 80x customer growth on Vercel's AI Gateway and about 27x daily token-volume growth—early evidence of real developer pull toward an open-weight frontier coder. ZCode plans start around $16/month.
What changed
Z.ai shipped ZCode, a dedicated agentic coding harness for its open-weight GLM-5.2 model, positioned for long, multi-step engineering tasks; GLM-5.2 is MIT-licensed with a 1M-token context and scores 62.1 on SWE-bench Pro. One week in, third-party platform data shows roughly 80x customer growth and ~27x daily token-volume growth for GLM-5.2 on Vercel's AI Gateway.
Why it matters
A capable open-weight coding model paired with a first-party agent harness gives teams a self-hostable alternative to closed agentic coders like Claude Code and Codex at a fraction of the cost, and the early Vercel adoption surge suggests the open-versus-closed shift is drawing real usage, not just benchmark attention. It also raises data-governance questions, since GLM-5.2 API calls are subject to Chinese data law.
Evidence excerpt
Z.ai shipped ZCode, the official harness for GLM-5.2 (launched July 2, 2026); GLM-5.2 scores 62.1 on SWE-bench Pro, ahead of GPT-5.5 at 58.6 and within four points of Claude Opus 4.8 at 66.0, and is MIT-licensed with a 1M-token context. In its first full week GLM-5.2 saw ~80x customer growth on Vercel with ~27x daily token-volume growth.