Can Desktop PC · RTX PRO 6000 Blackwell · 128 GB RAM run GLM-5.3-Flash (320B-A18B)?
96 GB VRAM + 128 GB RAM · 1792 GB/s memory bandwidth · GLM-5.3-Flash (320B-A18B) has 320B parameters (18B active per token).
Short answer
It loads, but it's borderline even for chat.
Best quantization for chat: Q4_K_M (186 GB of weights) on Ollama. Generation: 4.3–6 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 | 4.3–6 tok/s | ≈6 s | 8K | Details |
| Coding questionswith Continue | Borderline | Q4_K_M | 4.3–6 tok/s | ≈45 s | 16K | Details |
| Coding in a repositorywith Aider | Borderline | Q3_K_M | 7–10 tok/s | ≈90 s | 32K | Details |
| Agentic codingwith OpenCode | Borderline | Q3_K_M | 7–10 tok/s | ≈2 min | 32K | Details |
| Long documents (64K)with Open WebUI | Borderline | Q3_K_M | 7–9 tok/s | ≈8 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
Other models for Desktop PC · RTX PRO 6000 Blackwell · 128 GB RAM
The best-rated models for chat on this machine.
- Qwen3.6 35B-A3B260–360 tok/sExcellent
- Qwen3-Next 80B-A3B Instruct260–360 tok/sExcellent
- gpt-oss-120b≈180 tok/sExcellent
- Gemma 4 26B-A4B210–280 tok/sExcellent
- Qwen3.5 35B-A3B260–360 tok/sExcellent
- LFM2.5 24B-A2B530–720 tok/sExcellent
Cheapest machines for GLM-5.3-Flash (320B-A18B)
Hardware rated comfortable or better for chat, lowest price first.
Different tool, context size or workload?
These pages use typical settings. The full check models your exact setup.