The AI Bench
Zhipu AI (Z.ai) · MIT

GLM-4.7 Flash and GLM-4.5 Air

The Z.ai models you can run at home: a 30B-A3B for 24 GB machines and a 106B for 64–96 GB.

How much memory it needs

The model file, plus memory for the conversation, plus about 1 GB for the app. Longer conversations need more.

SizeVersionFileAt 8KAt 32KAt 128K
31B (3B active)Q4_K_M18.3 GB19.8 GB21.1 GB26.4 GB
Q8_031.8 GB33.3 GB34.6 GB39.9 GB
111B (12B active)Q4_K_M73.0 GB75.5 GB80.2 GB98.7 GB
Q8_0117 GB120 GB125 GB143 GB

Which computers run it

The best size for each machine at an 8K-token conversation. Speeds are estimates from memory bandwidth.

ComputerBest versionFitSpeed
Laptop or PC, no graphics card, 32 GB31B Q4_K_MRuns well15–30tok/sSmooth
RTX 4060 · 8 GB31B Q4_K_MSpills into system RAM18–35tok/sSmooth
RTX 3060 · 12 GB31B Q4_K_MSpills into system RAM25–50tok/sSmooth
RTX 5060 Ti · 16 GB31B Q4_K_MSpills into system RAM35–70tok/sFast
RTX 4090 · 24 GB31B Q4_K_MRuns well100–210tok/sFast
RTX 5090 · 32 GB31B Q4_K_MRuns great140–280tok/sFast
Mac with M4 · 16 GB31B Q4_K_MToo big—
Mac with M4 Pro · 48 GB31B Q8_0Runs well30–65tok/sFast
Mac with M5 Max · 128 GB111B Q4_K_MRuns great25–55tok/sFast
Ryzen AI Max+ 395 · 128 GB111B Q4_K_MRuns well10–20tok/sSmooth
DGX Spark · 128 GB111B Q4_K_MRuns great11–20tok/sSmooth
RTX PRO 6000 · 96 GB111B Q4_K_MRuns well55–110tok/sFast
Mac Studio with M5 Ultra · 256 GB111B Q8_0Runs great18–40tok/sSmooth
Mac Studio with M3 Ultra · 512 GB111B Q8_0Runs great13–30tok/sSmooth

What it is good and bad at

Good at

  • Coding and agents
  • MIT licence

Not good at

  • Newer GLM-5 models are too big for home machines
  • GLM-4.5 Air has no Ollama library entry

Start it

Install Ollama, then run one of these in a terminal. LM Studio users can search the same name in the app.

ollama run glm-4.7-flash:latest