Skip to content

Qwen3.6 35B A3B on M3 Ultra 96 GB: VRAM, speed and fit

Qwen3.6 35B A3B at 8-bit runs on the M3 Ultra 96 GB with 256K of context at about 85 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M3 Ultra 96 GB · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
65.4 GB
of 96.0 GB
Context
256K
auto
Speed
85 tok/s
46 when full
First reply
1.4 s
2K prompt

Where the memory goes

Model
35.6 GB
Context
5.1 GB
Overhead
805 MB
OS
24.0 GB
Free
30.6 GB
Memory
65.4 GB of 96.0 GB
Context
256K tokens · auto
Speed
85 tok/s · 46 when full
First reply
1.4 s for a 2K prompt
Running cost
$3.55/mo at home · $66/mo on a cloud 2× RTX A5000

Validation: Validated

What Qwen3.6 35B A3B on M3 Ultra 96 GB was checked against
CheckTargetResultStatus
Fits in GPU memory72.0 GB usable41.4 GB neededPass
Headroom5% free30.6 GB free (42%)Pass
Context for chat16K or more256KPass
Writing speed15 tok/s or more85 tok/sPass
First replyunder 3 s, 2K prompt1.4 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
Qwen3.6 35B A3B
Parameters
36B, 3B active
Precision
8-bit (Q8_0)
Context
256K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M3 Ultra 96 GB
GPU can use
72.0 GB of 96.0 GB

Running cost

At home
$3.55/mo
162 W at the wall
Renting instead
$66/mo
2× RTX A5000 at $0.54/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

Ollama
ollama run qwen3.6:35b-a3b
llama.cpp
llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF
vLLM
vllm serve Qwen/Qwen3.6-35B-A3B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.6-35B-A3B-4bit

Qwen3.6 35B A3B on similar machines

All machines: Qwen3.6 35B A3B VRAM requirements. Everything that runs here: what can the M3 Ultra 96 GB run?

Other models that run on M3 Ultra 96 GB

Questions

Can Qwen3.6 35B A3B run on M3 Ultra 96 GB?

Qwen3.6 35B A3B at 8-bit runs on the M3 Ultra 96 GB with 256K of context at about 85 tokens a second.

Report VR-1WCJJB6, calculated Nov 22, 2026 from published specs, not benchmarked.