PC for Qwen 3.6: 27B dense or 35B-A3B MoE, which choice?
Qwen 3.6 is the most talked-about open source model right now. But before choosing a PC to run it, you must answer one question: the dense 27B version, or the MoE 35B-A3B version? Everything follows from that.
Both belong to the same family. Both fit on a consumer card in 4-bit. Both are under Apache 2.0 license. But they behave very differently in use.
This guide answers the question first, then tells you the PC suited to your choice.
Qwen 3.6: what you need to know
Qwen 3.6 is Alibaba’s latest generation of open models. It’s a multimodal model (text, image, video), with hybrid reasoning, and an exceptional native context window reaching up to 1 million tokens.
Two open variants count for local use:
Qwen3.6-27B
Dense — code quality
- 27 billion parameters, all active
- Fits in about 16.8 GB in 4-bit
- Stable and predictable behavior
- Better on code and instruction following
- Includes vision
- Benefits from DFlash acceleration (recent NVIDIA)
Qwen3.6-35B-A3B
MoE — speed
- 35 billion total, 3 billion active per token
- Computational cost close to a 3B model
- Quality close to a 35B dense model
- Very fast: over 100 tokens/s reported on high-end GPU
- About 21 GB in 4-bit, requires 24 GB of VRAM
- Ideal for agents and fast tool use
Dense or MoE: how to choose
VRAM: what you really need to plan for
The figures below apply to a short to medium context. Warning: with extended context, the KV cache significantly increases the required memory.
| Variant | Quantization | VRAM (short context) | Recommended card |
|---|---|---|---|
| Qwen3.6-27B dense | Q4_K_M | about 16.8 GB | 16 GB tight, 24 GB comfortable |
| Qwen3.6-27B dense | Q5_K_M | about 20 GB | 24 GB |
| Qwen3.6-35B-A3B MoE | Q4_K_M | about 21 GB | 24 GB |
| Qwen3.6-35B-A3B MoE | Q5_K_M | about 26 GB | 32 GB |
| Either one, very long context | Q4 + quantized cache | +20 to 40 GB of KV cache | 32 GB and up |
Launch Qwen 3.6 in two minutes
The easiest way is through Ollama. Choose the variant according to your card:
# Dense 27B variant (16 GB and up) ollama run qwen3.6:27b # MoE 35B-A3B variant (24 GB and up) ollama run qwen3.6:35b-a3b # For agent or code use, llama.cpp is often used # with quantized KV cache for long context: llama-server -m qwen3.6-35b-a3b-Q4_K_M.gguf \ --cache-type-k q8_0 --cache-type-v q8_0 \ --ctx-size 65536 --n-gpu-layers 99
Which PC for Qwen 3.6
The choice of machine directly depends on the targeted variant. Here are our suitable workstations, assembled in Auriol (13390) and delivered throughout the EU, with Ollama and Open WebUI preinstalled on request.
Why run Qwen 3.6 locally
Beyond privacy, Qwen 3.6 locally offers concrete advantages for those who want a serious AI permanently.
- No recurring costs. No subscription, no token billing. Once the machine is purchased, usage is unlimited.
- Private data. Your prompts, code, and documents never leave your network.
- Top-tier quality. Qwen 3.6 rivals the best open models on code, reasoning, and agent tasks.
- Massive context. Up to 1 million tokens natively, to handle entire codebases or long documents.
- Apache 2.0 License. Free commercial use, no restrictions.
In brief
27B dense or 35B-A3B MoE?
Dense for code and reliable tool use on 16 GB. MoE for maximum speed on 24 GB and above.
What is the minimum VRAM?
16 GB for the 27B dense. 24 GB for the 35B-A3B. 32 GB for long context or Q5.
Is Qwen 3.6 free?
Yes, open source under Apache 2.0. You only pay for the hardware.
Can it be used for code and agents?
Yes, that's one of its strengths. For long agent loops, prefer the 27B dense, which is more consistent.
Do you need a powerful machine for the 1M token context?
Yes: the KV cache can add 20 to 40 GB. The GB10 mini-server with its 128 GB unified memory is the most comfortable in this regard.
