GLM-5.2

GLM-5.2 is Z.ai's flagship foundation model, released in June 2026 and — this is the part that matters — published under an MIT licence with no regional restrictions. That combination is rare: a model competitive with closed Western flagships on long-horizon coding work, whose weights anyone can download, self-host and use commercially without asking permission. Z.ai's own documentation describes a text-in, text-out model with a one-million-token context window and up to 128,000 tokens of output, supporting multiple thinking modes, streaming, function calling, context caching, structured JSON output and MCP tool integration. On Terminal-Bench 2.1 the documentation puts GLM-5.2 at 81.0 against 62.0 for its own predecessor GLM-5.1, and behind Claude Opus 4.8 at 85.0 — a candid comparison to include in your own product page, and a useful one, because it places the model precisely: at or near the top of the open-weight field, still short of the best closed model on that particular benchmark. Other widely quoted figures come from press coverage rather than Z.ai's specification: a 753-billion-parameter mixture-of-experts design with about 40 billion active parameters per token, an "IndexShare" attention scheme said to cut per-token FLOPs at long context, and scores such as 74.4% on FrontierSWE and 62.1 on SWE-bench Pro, with running costs reported at roughly one-sixth of GPT-5.5 for comparable coding tasks. Z.ai does not publish the parameter count in its model documentation, so we flag those numbers as reported rather than confirmed. The weights are distributed through Hugging Face and the model is served through Z.ai's own API and a long list of third-party coding tools.

▸Videos

No videos yet.

▸Reviews