GLM-4.7

GLM-4.7 closed Z.ai's fourth generation in December 2025, six weeks before GLM-5 opened the next one. It is not a legacy entry: the company still lists it in the current price table at $0.60 per million input tokens and $2.20 per million output, roughly a third of what the fifth-generation flagships cost, and it still ships the whole line — the paid GLM-4.7, the cheap GLM-4.7-FlashX at $0.07/$0.40, and GLM-4.7-Flash, which is free. The stated advance over GLM-4.6 is in three areas: coding, reasoning and tool use. Z.ai reports 73.8% on SWE-bench Verified (up 5.8 points), 66.7% on SWE-bench Multilingual (up 12.9), 41% on Terminal Bench 2.0 (up 16.5), 84.9 on LiveCodeBench V6, 84.7 on the τ²-Bench tool-calling benchmark and 42.8% on Humanity's Last Exam. The company describes the model as introducing a "think before acting" step inside coding tools such as Claude Code, Kilo Code, Cline and Roo Code. All of these are vendor figures, published on Z.ai's own product page. The weights are on Hugging Face under an MIT licence, and they settle what the documentation leaves out: the safetensors index totals 358.3 billion parameters — a little under half the size of GLM-5 — across 92 layers, with 160 routed experts plus one shared and eight active per token. The context window is 200,000 tokens with up to 128,000 of output, and the feature set is the familiar one: thinking modes, streaming, function calling, context caching, structured output and MCP tools.

▸Videos

No videos yet.

▸Reviews