No — not on this device
Qwen3.8 2.4T-A95B at Q4_K_M needs 1376.7 GB against 24.0 GB usable, and the shortfall of 1352.7 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 open Qwen-Max-class flagship. Listed as the honest ceiling: nothing short of a rack runs it.
What hardware do I need for Qwen3.8 2.4T-A95B? →
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 | 2420.4 GB | 2421.7 GB | — | ~0.3 | −0.1% ppl | 2397.7 GB over |
| Q6_K | 1868.0 GB | 1869.3 GB | — | ~0.4 | −0.4% ppl | 1845.3 GB over |
| Q5_K_M | 1614.5 GB | 1615.9 GB | — | ~0.5 | −0.8% ppl | 1591.9 GB over |
| Q4_K_M | 1375.4 GB | 1376.7 GB | — | ~0.6 | −1.9% ppl | 1352.7 GB over |
| Q3_K_M | 1113.4 GB | 1114.7 GB | — | ~0.7 | −5.4% ppl | 1090.7 GB over |
| Q2_K | 953.9 GB | 955.2 GB | — | ~0.8 | −15% ppl | 931.2 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 23 of its 92 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/Qwen3.8-2.4T-A95B-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.