Alibaba · 35B (3B active) · Mixture of Experts
High-sparsity MoE variant of Qwen 3.6 tuned for agentic coding workflows
0 downloads
0 likes
2026-08 256K context
Use Cases
chat code reasoning
Mixture of Experts
Total experts: 256
Active experts: 8
Active params: 3.0B
| Quant | Bits | VRAM | Quality | Status |
|---|---|---|---|---|
| Q2_K | 2 | 11.7 GB | low | — |
| Q3_K_M | 3 | 16.2 GB | moderate | — |
| Q4_K_M | 4 | 18.4 GB | good | — |
| Q5_K_M | 5 | 22.9 GB | good | — |
| Q6_K | 6 | 27.4 GB | excellent | — |
| Q8_0 | 8 | 36.4 GB | excellent | — |
| F16 | 16 | 72.2 GB | lossless | — |
GPUs that run Qwen 3.6 35B-A3B
Qwen 3.6 35B-A3B needs about 18.4 GB of VRAM at Q4_K_M, so it runs comfortably on a GPU with 24 GB of VRAM or more (35 of the GPUs we track). Common cards, smallest first:
| VRAM | GPUs |
|---|---|
| 24 GB | RTX 4090 , RX 7900 XTX , RTX 3090 Ti |
| 32 GB | RTX 5090 , RTX 5000 Ada , Tesla V100 32GB |
About this model
High-sparsity MoE variant of Qwen 3.6 tuned for agentic coding workflows