GLM-5

GLM-5 opened Z.ai's fifth generation in February 2026 and set the pattern the whole family follows: a foundation model built for agentic engineering, published under an MIT licence, with weights anyone can download and use commercially. Z.ai's documentation positions it for long-range agent tasks and complex system engineering, and claims state-of-the-art results among open-source models on coding and agent work, with real-world usability approaching Claude Opus 4.5. The published specification is modest in scope: text in, text out, a 200,000-token context window and up to 128,000 tokens of output, with thinking modes, streaming, function calling, context caching, structured JSON output and MCP tool integration. Z.ai charges $1.00 per million input tokens and $3.20 per million output — the cheapest tier in the GLM-5 line, since every later version costs more. One detail is worth spelling out, because it recurs across this family: Z.ai does not state a parameter count anywhere in its model documentation, and the figure quoted in press coverage of later versions is therefore "reported" rather than confirmed. For GLM-5 it can be confirmed, because the company shipped the weights: the safetensors index on Hugging Face totals 753.9 billion parameters, and the configuration file describes a mixture-of-experts network of 78 layers with 256 routed experts plus one shared expert, eight of which are active per token, using a sparse-attention indexer. That is the same order of size the press attributes to GLM-5.2 — but here it comes from the model file, not from a briefing.

▸Videos

No videos yet.

▸Reviews