# Qwen3.6-35B-A3B

> OnPrem World · https://onpremworld.com/en/models/qwen3.6-35b-a3b · Last updated: 2026-10-14 · CC BY 4.0

The April 2026 refresh of Qwen3.5-35B-A3B: the same 3B-active MoE design and memory needs, aimed at coding and agents.

|  |  |
|---|---|
| Developer | Alibaba Qwen |
| Parameters | 35B (3B active, MoE) |
| Native context | 262,144 tokens |
| Licence | Apache-2.0 |
| Commercial use | Yes |
| KV cache | 10 of 40 layers (the rest are linear attention with a fixed-size state) |
| Released | 2026-04 |

```
ollama run qwen3.6:35b
```

## Memory needed

Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.

| Quantization | 4K | 8K | 32K |
|---|---|---|---|
| Q4_K_M | 21.3 GiB | 21.4 GiB | 21.9 GiB |
| Q5_K_M | 24.9 GiB | 25.0 GiB | 25.5 GiB |
| Q6_K | 28.8 GiB | 28.9 GiB | 29.3 GiB |
| Q8_0 | 36.9 GiB | 37.0 GiB | 37.5 GiB |
| FP16 | 69.0 GiB | 69.1 GiB | 69.6 GiB |

## Which hardware runs it (8K)

| Hardware | Memory | Best quant | tok/s (est.) |
|---|---|---|---|
| GeForce RTX 5090 | 32 GB | Q6_K | 508 |
| GeForce RTX 4090 | 24 GB | Q4_K_M | 388 |
| Radeon RX 7900 XTX | 24 GB | Q4_K_M | 369 |
| GeForce RTX 3090 | 24 GB | Q4_K_M | 360 |
| RTX 6000 Ada Generation | 48 GB | Q8_0 | 211 |
| RTX PRO 6000 Blackwell Workstation Edition | 96 GB | FP16 | 209 |
| Radeon AI PRO R9700 | 32 GB | Q6_K | 181 |
| Arc Pro B60 | 24 GB | Q4_K_M | 176 |
| Arc Pro B70 | 32 GB | Q6_K | 172 |
| RTX A6000 | 48 GB | Q8_0 | 169 |
| RTX PRO 5000 72GB Blackwell | 72 GB | FP16 | 157 |
| M5 Max (18-core CPU / 32-core GPU) | 36 GB | Q5_K_M | 140 |
| M5 Ultra | 512 GB | FP16 | 130 |
| M4 Max (14-core CPU / 32-core GPU) | 36 GB | Q5_K_M | 125 |
| M3 Ultra | 512 GB | FP16 | 89 |
| M5 Max (18-core CPU / 40-core GPU) | 128 GB | FP16 | 67 |
| M5 Pro | 64 GB | Q8_0 | 63 |
| M6 (24GB/32GB configurations, 170 GB/s) | 32 GB | Q4_K_M | 61 |
| M4 Max (16-core CPU / 40-core GPU) | 128 GB | FP16 | 59 |
| M4 Pro | 64 GB | Q8_0 | 56 |
| M5 | 32 GB | Q4_K_M | 55 |
| RTX Spark (N1X) laptops & compact desktops | 128 GB | FP16 | 30 |
| DGX Spark | 128 GB | FP16 | 27 |
| Ryzen AI Max+ PRO 495 (Gorgon Halo) systems | 192 GB | FP16 | 27 |
| Ryzen AI Max+ 395 (Strix Halo) systems | 128 GB | FP16 | 26 |
| GeForce RTX 3060 (12 GB) | 12 GB | Doesn't fit | – |
| GeForce RTX 4060 Ti 16GB | 16 GB | Doesn't fit | – |
| GeForce RTX 4080 SUPER | 16 GB | Doesn't fit | – |
| GeForce RTX 5060 Ti 16GB | 16 GB | Doesn't fit | – |
| GeForce RTX 5070 | 12 GB | Doesn't fit | – |
| GeForce RTX 5070 Ti | 16 GB | Doesn't fit | – |
| GeForce RTX 5080 | 16 GB | Doesn't fit | – |
| Radeon RX 7900 XT | 20 GB | Doesn't fit | – |
| Radeon RX 9060 XT 16GB | 16 GB | Doesn't fit | – |
| Radeon RX 9070 XT | 16 GB | Doesn't fit | – |
| Arc B580 | 12 GB | Doesn't fit | – |
| M6 (16GB configuration, 153 GB/s) | 16 GB | Doesn't fit | – |

## Frequently asked questions

### How much VRAM does Qwen3.6-35B-A3B need?

With an 8K context, about 21.4 GiB at Q4_K_M and 37.0 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.

### Can Qwen3.6-35B-A3B run on an GeForce RTX 4090?

Yes. Qwen3.6-35B-A3B fits on the GeForce RTX 4090 (24 GB) at Q4_K_M with an 8K context, using about 21.4 GiB, at an estimated 388 tokens/s.

### Can I run Qwen3.6-35B-A3B on a Mac?

Yes. Qwen3.6-35B-A3B fits on the M3 Ultra (96 GB) at FP16 with an 8K context, using about 69.1 GiB, at an estimated 89 tokens/s.

### Can Qwen3.6-35B-A3B be used commercially?

Qwen3.6-35B-A3B is released under Apache-2.0. The licence allows commercial use (read its conditions).

## Sources

- https://github.com/QwenLM/Qwen3.8/blob/main/README.md
- https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/main/LICENSE
- https://huggingface.co/Qwen/Qwen3.6-35B-A3B
- https://huggingface.co/Qwen/Qwen3.6-35B-A3B/resolve/main/config.json
- https://huggingface.co/api/models/Qwen/Qwen3.6-35B-A3B?expand[]=safetensors
- https://ollama.com/library/qwen3.6/tags