Skip to content

GLM 4.5 Air on M3 Ultra 96 GB: VRAM, speed and fit

GLM 4.5 Air at 4-bit runs on the M3 Ultra 96 GB with 28K of context at about 43 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M3 Ultra 96 GB · 4-bit (Q4_K_M) · Ollama / LM Studio · Chat

GPU memory
92.1 GB
of 96.0 GB
Context
28K
auto
Speed
43 tok/s
30 when full
First reply
9.0 s
2K prompt

Where the memory goes

Model
62.4 GB
Context
5.0 GB
Overhead
680 MB
OS
24.0 GB
Free
3.9 GB
Memory
92.1 GB of 96.0 GB
Context
28K tokens · auto
Speed
43 tok/s · 30 when full
First reply
9.0 s for a 2K prompt
Running cost
$3.55/mo at home · $122/mo on a cloud 4× RTX A4500

Validation: Not validated

What GLM 4.5 Air on M3 Ultra 96 GB was checked against
CheckTargetResultStatus
Fits in GPU memory72.0 GB usable68.1 GB neededPass
Headroom5% free3.9 GB free (5%)Pass
Context for chat16K or more28KPass
Writing speed15 tok/s or more43 tok/sPass
First replyunder 3 s, 2K prompt9.0 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
GLM 4.5 Air
Parameters
110B, 15B active
Precision
4-bit (Q4_K_M)
Context
28K, 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

Worth trying

  • 8-bit cache: Room for 56K context

Running cost

At home
$3.55/mo
162 W at the wall
Renting instead
$122/mo
4× RTX A4500 at $1.00/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/GLM-4.5-Air-GGUF
vLLM
vllm serve zai-org/GLM-4.5-Air
MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-4.5-Air-4bit

GLM 4.5 Air on similar machines

All machines: GLM 4.5 Air 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 GLM 4.5 Air run on M3 Ultra 96 GB?

GLM 4.5 Air at 4-bit runs on the M3 Ultra 96 GB with 28K of context at about 43 tokens a second.

Report VR-1H1KK12, calculated Nov 22, 2026 from published specs, not benchmarked.