Qwen3.5 397B-A17B
Qwen/Qwen3.5-397B-A17B
Parameters
397B
Context
256K
Downloads
315.7K
Likes
1.6K
Architecture
moe_transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 940.4 GB RAM 1410.6 · Disk 1192.6 | 998.7 GB RAM 1498.0 · Disk 1218.6 | 1138.1 GB RAM 1707.1 · Disk 1257.6 |
| FP88bit | 472.7 GB RAM 709.0 · Disk 615.8 | 509.7 GB RAM 764.5 · Disk 641.8 | 627.8 GB RAM 941.8 · Disk 680.8 |
| INT88bit | 472.7 GB RAM 709.0 · Disk 615.8 | 509.7 GB RAM 764.5 · Disk 641.8 | 627.8 GB RAM 941.8 · Disk 680.8 |
| AWQ4bit | 247.6 GB RAM 371.4 · Disk 338.2 | 274.4 GB RAM 411.5 · Disk 364.2 | 382.3 GB RAM 573.4 · Disk 403.2 |
| GPTQ4bit | 250.5 GB RAM 375.8 · Disk 341.8 | 277.4 GB RAM 416.1 · Disk 367.8 | 385.5 GB RAM 578.2 · Disk 406.8 |
| GGUF4bit | 256.4 GB RAM 384.5 · Disk 349.0 | 283.5 GB RAM 425.3 · Disk 375.0 | 391.9 GB RAM 587.8 · Disk 414.0 |
GPU Recommendations
Best ValueBEST
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 80.1 tok/s
Effective production throughput ≈ 56.1 tok/s
score 0.2846
Cheapest
8× RTX A6000
384 GB VRAM
$1,806/mo
≈ 12.8 tok/s
Effective production throughput ≈ 9 tok/s
score 0.5305
Performance
4× H200 SXM
564 GB VRAM
$10,220/mo
≈ 60.1 tok/s
Effective production throughput ≈ 42.1 tok/s
score 0.3991
Min Complexity
2× H200 SXM
282 GB VRAM
$5,110/mo
≈ 40 tok/s
Effective production throughput ≈ 28 tok/s
score 0.5426
💬 Community — real deployment experience
0 Articles · 0 Benchmarks