Qwen3.8 family: which size to run
2 sizes, 27B–2.4T parameters · Alibaba
Sizes
| Model | Architecture | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen3.8 27B | 27B | Dense | 262,144 | 75.3 | 29 | 23 | 16.5† | 16.1 |
| Qwen3.8 Max | 2.4T | MoE · 95B active | 262,144 | 78.5 | 1,808 | 1,802 | No GGUF Q4_K_M file (plain or UD-) in the repository collected | No MLX repository collected |
Each size is its own model, with its own scores and memory needs. Q4 estimates are planning figures, not measurements; file sizes are published. The Mac or PC column is for a 32 GB or larger machine (8 GB reserve); machines under 32 GB reserve 25% of their memory (2 to 8 GB), so they need less. The publisher lists 27B, 2.4T-A95B.
† Unsloth's dynamic UD-Q4_K_M file, shown when the repository has no plain Q4_K_M file (the same rule as the model pages); the cell's title names the file.
How to read this table
Q4 means compressed to about 4 bits per weight. GGUF (llama.cpp, Ollama, LM Studio) and MLX (Apple silicon) are file formats; the sizes shown are the published 4-bit files. A dash means no value is published or collected for that size; the reason is in the cell's accessible text.
Which size fits my machine
The largest size whose Q4 planning estimate fits what each machine offers. A planning estimate, not a measurement and not a speed. Smaller machines are left out; the largest one that fits no size is kept to show the limit.
| Machine | Largest Qwen3.8 size with enough memory | ||
|---|---|---|---|
| Mac mini M4 · 24 GB unified | 24 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| Mac mini M4 · 32 GB unified | 32 | Qwen3.8 27B | 29 |
| Mac mini M5 Pro · 64 GB unified | 56 | Qwen3.8 27B | 29 |
| Mac Studio M5 Max · 128 GB unified | 104 | Qwen3.8 27B | 29 |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unified | 392 | Qwen3.8 27B | 29 |
| RTX 5080 · 16 GB VRAM | 16 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| RTX 4090 · 24 GB VRAM | 24 | Qwen3.8 27B | 23 |
| RTX 5090 · 32 GB VRAM | 32 | Qwen3.8 27B | 23 |
| DGX H100 · 8 × 80 GB | 640 | Qwen3.8 27B | 37 |
A dash means no size of this family fits that machine. The requirement is the size's Q4 estimate with that machine's own reserve: a graphics card reserves 2 GB, as in the sizes table; a Mac or PC reserves 25% of its memory (2 to 8 GB), so below 32 GB its figure is lower than the sizes table's. A server profile reserves 2 GB per graphics card, so its figure is higher.
Smallest machine for each size
The profile with the least memory that still has enough for the size's Q4 estimate, from the hardware catalog.
| Size | Smallest Mac or PC-memory profile | Smallest graphics-card profile | Server-class profile |
|---|---|---|---|
| Qwen3.8 27B | Mac mini 2024 · M4 · 32 GB | RTX 3090 · 24 GB | DGX H100 · 8 × 80 GB · 640 GB |
| Qwen3.8 Max | No Mac, laptop or PC profile in the catalog has enough memory for this size | No graphics-card profile in the catalog has enough memory for this size | Even the largest profile in the catalog (DGX H100 · 8 × 80 GB, 640 GB aggregated) is below this planning estimate |
A dash means no profile of that kind in the catalog has enough memory for the size.
Memory by file precision
The same planning formula at three precisions (0.625, 1.125 and 2 bytes per parameter, plus working room and the system reserve). Estimates, not measurements.
| Size | ||||||
|---|---|---|---|---|---|---|
| Qwen3.8 27B | 29 | 23 | 45 | 39 | 73 | 67 |
| Qwen3.8 Max | 1,808 | 1,802 | 3,248 | 3,242 | 5,768 | 5,762 |
Family tree
Family, sizes and, for each size, the two 4-bit files the tables use (GGUF Q4_K_M and MLX 4-bit) when one was collected. Other precisions are not drawn here.
The same tree as a list:
- Qwen3.8 · Alibaba
- 27B
- GGUF UD-Q4_K_M · 16.5 GB · unsloth/Qwen3.8-27B-GGUF, read Oct 23, 2026
- MLX 4-bit · 16.1 GB · lmstudio-community/Qwen3.8-27B-MLX-4bit, read Oct 23, 2026
- Max 2.4T A95B · 95B active
- No Q4_K_M or 4-bit file collected
- 27B
Learn more
See Qwen3.8 27B
The smallest size of the family: its scores, memory estimate and the machines it runs on.
Sources
Where the family and its sizes are named, and where each number comes from.
- QwenLM/Qwen3.5 README · “2026-08-14: Qwen3.8-27B is now available on Hugging Face Hub” · read Oct 17, 2026
- QwenLM/Qwen3.5 README · “2026-08-12: Qwen3.8-2.4T-A95B is now available on Hugging Face Hub” · read Oct 17, 2026
- Qwen/Qwen3.8-Flash-Next model card · “This experimental preview of the architecture that will underpin Qwen4” · read Oct 17, 2026
- abacusai/Smaug-Mini model card · “Smaug-Mini is an agentic finetune of Qwen3.8-27B, finetuned by Abacus.AI.” · read Oct 17, 2026
- Qwen3.8 27B model card
- Qwen3.8 Max model card
- LiveBench 2026-06-25 · the only quality source on this page; a dash means the release does not include the model · snapshot collected Sep 26, 2026
- Published file sizes: Hugging Face repositories, read Oct 23, 2026
- Methodology · how the Q4 planning estimate is computed · Learn