Can Desktop PC · Radeon RX 7900 XTX · 64 GB RAM run GLM-4.5-Air (106B-A12B)?
24 GB VRAM + 64 GB RAM · 960 GB/s memory bandwidth · GLM-4.5-Air (106B-A12B) has 110.5B parameters (12B active per token).
Short answer
It loads, but it's borderline even for chat.
Best quantization for chat: Q4_K_M (68 GB of weights) on Lemonade. Generation: 5–7 tok/s, leaving about 0.0 GB free for your other apps.
How it rates for each workload
Same model and machine, different jobs. Agents and long documents send far bigger prompts than chat, so they need much more speed and memory.
| Workload | Verdict | Quant | Generation | First reply | Context | |
|---|---|---|---|---|---|---|
| Chatwith Open WebUI | Borderline | Q4_K_M | 5–7 tok/s | ≈4 s | 8K | Details |
| Coding questionswith Continue | Borderline | Q4_K_M | 4.6–6 tok/s | ≈25 s | 16K | Details |
| Coding in a repositorywith Aider | Borderline | Q3_K_M | 5–7 tok/s | ≈60 s | 32K | Details |
| Agentic codingwith OpenCode | Borderline | Q3_K_M | 4.7–6 tok/s | ≈85 s | 32K | Details |
| Long documents (64K)with Open WebUI | Borderline | Q3_K_M | 2.8–3.8 tok/s | ≈6 min | 64K | Details |
Speeds are estimates calibrated against measured benchmarks on similar hardware. “First reply” is the time to read the workload’s opening prompt. How we calculate
Get it running
Copy-paste commands for the Q4_K_M build with a 8K context. Ollama and llama.cpp download the file straight from Hugging Face.
ollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q4_K_M
This build is split into 2 files, which Ollama can't import from Hugging Face. Use llama.cpp or LM Studio instead.
Other models for Desktop PC · Radeon RX 7900 XTX · 64 GB RAM
The best-rated models for chat on this machine.
- Qwen3.6 35B-A3B240–320 tok/sExcellent
- Gemma 4 26B-A4B160–220 tok/sExcellent
- LFM2.5 24B-A2B350–480 tok/sExcellent
- gpt-oss-20b210–280 tok/sExcellent
- Qwen3.5 35B-A3B240–320 tok/sExcellent
- GLM-4.7-Flash (30B-A3B)240–320 tok/sExcellent
Cheapest machines for GLM-4.5-Air (106B-A12B)
Hardware rated comfortable or better for chat, lowest price first.
- Ryzen AI Max+ 395 mini PC / Framework Desktop · 128 GB$2,000 · 17–25 tok/sExcellent
- Mac Studio M2 Max 96 GB$3,799 · 25–35 tok/sExcellent
- MacBook Pro M3 Max 96 GB$3,999 · 19–25 tok/sExcellent
- Mac Studio M5 Max 128 GB$4,099 · 25–45 tok/sExcellent
- Mac Studio M4 Max 128 GB$4,099 · 25–35 tok/sExcellent
- MacBook Pro M2 Max 96 GB$5,099 · 25–35 tok/sExcellent
Same chip, other memory: 32 GB
Different tool, context size or workload?
These pages use typical settings. The full check models your exact setup.