Qwen3 235B A22B
Qwen/Qwen3-235B-A22B
MoE variant, 235B total / 22B active. Needs multi-GPU or 141GB+ VRAM.
Parameters
235B
Context
256K
Downloads
460K
Likes
3.1K
Architecture
moe_transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 558.7 GB RAM 838.0 · Disk 721.9 | 599.6 GB RAM 899.4 · Disk 747.9 | 721.7 GB RAM 1082.5 · Disk 786.9 |
| FP88bit | 281.8 GB RAM 422.7 · Disk 380.4 | 310.1 GB RAM 465.2 · Disk 406.4 | 419.6 GB RAM 629.4 · Disk 445.4 |
| INT88bit | 281.8 GB RAM 422.7 · Disk 380.4 | 310.1 GB RAM 465.2 · Disk 406.4 | 419.6 GB RAM 629.4 · Disk 445.4 |
| AWQ4bit | 148.6 GB RAM 222.9 · Disk 216.1 | 170.9 GB RAM 256.3 · Disk 242.1 | 274.3 GB RAM 411.4 · Disk 281.1 |
| GPTQ4bit | 150.3 GB RAM 225.5 · Disk 218.3 | 172.7 GB RAM 259.0 · Disk 244.3 | 276.2 GB RAM 414.3 · Disk 283.3 |
| GGUF4bit | 153.8 GB RAM 230.6 · Disk 222.5 | 176.3 GB RAM 264.4 · Disk 248.5 | 279.9 GB RAM 419.9 · Disk 287.5 |
GPU Recommendations
Best ValueBEST
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 135.3 tok/s
Effective production throughput ≈ 94.7 tok/s
score 0.2601
Cheapest
8× RTX 3090
192 GB VRAM
$630/mo
≈ 26.4 tok/s
Effective production throughput ≈ 18.5 tok/s
score 0.531
Performance
4× H200 SXM
564 GB VRAM
$10,220/mo
≈ 101.5 tok/s
Effective production throughput ≈ 71.1 tok/s
score 0.3501
Min Complexity
1× B200
180 GB VRAM
$3,013/mo
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.5741
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
B200
180 GB VRAM
$3,013
/mo · vast
B200 SXM
180 GB VRAM
—