No — not on this device
GLM-5.3-Flash 320B-A18B at Q4_K_M needs 181.2 GB against 36.0 GB usable, and the shortfall of 145.2 GB is more than 32 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
The first natively multimodal GLM-5 and the first hybrid: 34 linear-attention blocks and 11 sparse-attention blocks with a 512-wide latent cache, so a 1M window stays affordable. Z.ai says it beats GLM-5.2 at 18B active; MIT.
What hardware do I need for GLM-5.3-Flash 320B-A18B? →
Fits instead: GLM-4.7-Flash 30B-A3B (18.6 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 317.6 GB | 318.3 GB | — | ~4.0 | −0.1% ppl | 282.3 GB over |
| Q6_K | 245.1 GB | 245.8 GB | — | ~5.2 | −0.4% ppl | 209.8 GB over |
| Q5_K_M | 211.9 GB | 212.6 GB | — | ~6.0 | −0.8% ppl | 176.6 GB over |
| Q4_K_M | 180.5 GB | 181.2 GB | — | ~7.0 | −1.9% ppl | 145.2 GB over |
| Q3_K_M | 146.1 GB | 146.8 GB | — | ~8.7 | −5.4% ppl | 110.8 GB over |
| Q2_K | 125.2 GB | 125.9 GB | — | ~10 | −15% ppl | 89.9 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 11 of its 45 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/GLM-5.3-Flash-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.