Skip to content

Qwen3 235B A22B on M3 Ultra 256 GB: VRAM, speed and fit

Qwen3 235B A22B at 5-bit runs on the M3 Ultra 256 GB with 40K of context at about 23 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M3 Ultra 256 GB · 5-bit (Q5_K_M) · Ollama / LM Studio · Chat

GPU memory
228 GB
of 256 GB
Context
40K
auto
Speed
23 tok/s
17 when full
First reply
14 s
2K prompt

Where the memory goes

Model
156 GB
Context
7.3 GB
Overhead
681 MB
OS
64.0 GB
Free
28.3 GB
Memory
228 GB of 256 GB
Context
40K tokens · auto
Speed
23 tok/s · 17 when full
First reply
14 s for a 2K prompt
Running cost
$3.55/mo at home · $287/mo on a cloud 4× A40

Validation: Not validated

What Qwen3 235B A22B on M3 Ultra 256 GB was checked against
CheckTargetResultStatus
Fits in GPU memory192 GB usable164 GB neededPass
Headroom5% free28.3 GB free (15%)Pass
Context for chat16K or more40KPass
Writing speed15 tok/s or more23 tok/sPass
First replyunder 3 s, 2K prompt14 sFails

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 235B A22B
Parameters
235B, 22B active
Precision
5-bit (Q5_K_M)
Context
40K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M3 Ultra 256 GB
GPU can use
192 GB of 256 GB

Running cost

At home
$3.55/mo
162 W at the wall
Renting instead
$287/mo
4× A40 at $2.36/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

llama.cpp
llama-server -hf unsloth/Qwen3-235B-A22B-GGUF
vLLM
vllm serve Qwen/Qwen3-235B-A22B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3-235B-A22B-4bit

Qwen3 235B A22B on similar machines

All machines: Qwen3 235B A22B VRAM requirements. Everything that runs here: what can the M3 Ultra 256 GB run?

Other models that run on M3 Ultra 256 GB

Questions

Can Qwen3 235B A22B run on M3 Ultra 256 GB?

Qwen3 235B A22B at 5-bit runs on the M3 Ultra 256 GB with 40K of context at about 23 tokens a second.

Report VR-12S5IRL, calculated Nov 22, 2026 from published specs, not benchmarked.