text model · Qwen3 · macOS

Can I run Qwen3 235B A22B on Apple M5 Max (128GB)?

Compatibility verdict VRAM threshold engine

No, not enough memory would not load

No. Qwen3 235B A22B needs ~136.9 GB even at Q4_K_M, but Apple M5 Max (128GB) only has ~96 GB usable.

Needs ~136.9 GB Device usable ~96 GB

Needs ~136.9 GB even at Q4_K_M, but only ~96 GB is usable.

That figure is at a 4k context and moves about ±15% as context length changes.

The gap is about 40.9 GB: Qwen3 235B A22B needs roughly 136.9 GB at Q4_K_M and Apple M5 Max (128GB) leaves only about 96 GB usable for a model. The lightest tracked hardware that runs Qwen3 235B A22B is the Apple M3 Ultra (256GB) at 256 GB. See Qwen3 235B A22B on Apple M3 Ultra (256GB).

Q4_K_M needed: ~136.9 GB
Usable on device: ~96 GB
Device memory: 128 GB

Share on X Share on Reddit

Which quant fits

Quant ladder vs ~96 GB usable

Q2_K

~102.9 GB

Q3_K_M

~119.4 GB

Q4_K_M

~136.9 GB

Q5_K_M

~171.9 GB

Q6_K

~197.2 GB

Q8_0

~237.3 GB

FP16

~474.5 GB

The line marks Apple M5 Max (128GB)'s ~96 GB budget; rungs past it are too large.

Model Qwen3

Parameters: 235B (MoE, 22B active)
Q4_K_M size: 132.39 GB
Q8_0 size: 232.77 GB
Context: 128k
Ollama tag: qwen3:235b

Full Qwen3 235B A22B requirements →

Device macOS

Memory: 128 GB unified
Usable for weights: ~96 GB
Best runtime: MLX direct / Ollama (MLX backend)

Best models for Apple M5 Max (128GB) →

What you can run instead

Run Qwen3 235B A22B on other hardware

Apple M3 Ultra (256GB)

FAQ

Can Apple M5 Max (128GB) run Qwen3 235B A22B?

No. Qwen3 235B A22B needs ~136.9 GB even at Q4_K_M, but Apple M5 Max (128GB) only has ~96 GB usable.

How much memory does Qwen3 235B A22B need?

Apple M5 Max (128GB) does not have enough memory. At Q4_K_M the weights are ~132.39 GB; with KV cache and runtime overhead, budget ~136.9 GB at a 4k context. It is a Mixture-of-Experts model (235B total / 22B active), so all experts must stay in memory; memory tracks total params, not active params.

What is the best tool to run Qwen3 235B A22B on macOS?

LM Studio for a simple setup; mlx-lm for the most speed. vLLM is NOT a Mac tool, it is a CUDA/Linux serving engine. Unified memory is not a fixed VRAM slice; ~70% is usable for weights.

Embed this

A live badge for your model card or README, updated as the data is.

Markdown

[![Qwen3 235B A22B on Apple M5 Max (128GB)](https://localmodel.run/badge/qwen3-235b-a22b/apple-m5-max-128gb.svg)](https://localmodel.run/can-i-run/qwen3-235b-a22b/apple-m5-max-128gb)

Sources

Weights are measured from GGUF files; KV cache and overhead are computed, so totals can vary ~15% with context and runtime. Any tok/s is a bandwidth estimate. See methodology.