Autor: Jakub Rusinowski · Ostatnia aktualizacja: 15 sierpnia 2026
Model MoE Zhipu AI z lutego 2026. GLM-4.7 używa architektury MoE 32B aktywnych i na benchmarkach halucynacji dorównuje Claude Opus 4.5. Wariant GLM-Z1 zorientowany na rozumowanie osiąga wyniki bliskie GPT-5.2. Mocna obsługa wielojęzykowa z głęboką znajomością języka chińskiego.
| Licence | What it permits | Applies to |
|---|---|---|
Apache-2.0 | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | GLM-4.7 9B, GLM-Z1 32B (Reasoning) |
MIT | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | GLM-4.7-Flash 30B-A3B |
| GLM-4.7 9B | Min 6 GB VRAM · Q4_K_M · 128,000 ctx · ollama run glm4:9b |
| GLM-Z1 32B (Reasoning) | Min 20 GB VRAM · Q4_K_M · 128,000 ctx · ollama run glm-z1:32b |
| GLM-4.7-Flash 30B-A3B | Min 19 GB VRAM · Q4_K_M · 198,000 ctx · ollama run glm-4.7-flash |
The cheapest GPU that runs GLM-4.7 / GLM-Z1 locally (min 6 GB VRAM) is the Intel Arc B570 (10 GB).
Install Ollama then run: ollama run glm4:9b
Minimum VRAM: 6 GB. For best results use Q4_K_M quantization.
GLM-4.7 / GLM-Z1 needs about 6 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: GLM-4.7 9B (6 GB, Q4_K_M); GLM-Z1 32B (Reasoning) (20 GB, Q4_K_M); GLM-4.7-Flash 30B-A3B (19 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Yes — GLM-4.7 / GLM-Z1 runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
Q4_K_M is the best balance of quality and VRAM for GLM-4.7 / GLM-Z1 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run glm4:9b. This downloads GLM-4.7 / GLM-Z1 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.