No — not on this device
Qwen3.8 2.4T-A95B at Q4_K_M needs 1376.7 GB against 14.4 GB usable, and the shortfall of 1362.3 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.
Does not fit
8K context
Q4_K_M · 1375.4 GB
Qwen3.8-Max License
Released 12 Aug 2026
New this month
Not in the Ollama library
The first open Qwen-Max-class flagship. Listed as the honest ceiling: nothing short of a rack runs it.
The VRAM budget
weights 1375.4 GB
Weights 1375.4 GB
KV cache @ 8K 0.72 GB
Runtime overhead 0.6 GB
Over budget 1362.3 GB past 14.4 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 2420.4 GB | 2421.7 GB | — | ~0.1 | −0.1% ppl | 2407.3 GB over |
| Q6_K | 1868.0 GB | 1869.3 GB | — | ~0.2 | −0.4% ppl | 1854.9 GB over |
| Q5_K_M | 1614.5 GB | 1615.9 GB | — | ~0.2 | −0.8% ppl | 1601.5 GB over |
| Q4_K_M | 1375.4 GB | 1376.7 GB | — | ~0.3 | −1.9% ppl | 1362.3 GB over |
| Q3_K_M | 1113.4 GB | 1114.7 GB | — | ~0.3 | −5.4% ppl | 1100.3 GB over |
| Q2_K | 953.9 GB | 955.2 GB | — | ~0.4 | −15% ppl | 940.8 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
$ llama-server \
-hf Qwen/Qwen3.8-2.4T-A95B:Q4_K_M \
-c 8192 -ngl 0
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 1375.4 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.