No — not on this device
Qwen3.8-Flash-Next 180B-A6B at Q4_K_M needs 102.0 GB against 96.0 GB usable, and the shortfall of 6.0 GB is more than 128 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
The open preview of the Qwen4 architecture: a 125B-A6B hybrid (Gated DeltaNet + sparse attention, KV cache on 12 of 48 blocks) plus a 51B n-gram embedding and a 4B draft head — 180B on disk, 6B active. The hosted "Qwen3.8-Flash" is this model with a 1M window.
What hardware do I need for Qwen3.8-Flash-Next 180B-A6B? →
Fits instead: Qwen3.8 27B (16.7 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 178.1 GB | 178.9 GB | — | ~35 | −0.1% ppl | 82.9 GB over |
| Q6_K | 137.5 GB | 138.3 GB | — | ~46 | −0.4% ppl | 42.3 GB over |
| Q5_K_M | 118.8 GB | 119.6 GB | — | ~53 | −0.8% ppl | 23.6 GB over |
| Q4_K_M | 101.2 GB | 102.0 GB | — | ~62 | −1.9% ppl | 6.0 GB over |
| Q3_K_M | 81.9 GB | 82.7 GB | 256K | 76 | −5.4% ppl | Long context |
| Q2_K | 70.2 GB | 71.0 GB | 256K | 89 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 12 of its 48 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/Qwen3.8-Flash-Next-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.