Skip to content

GLM 4.5 Air on 4× RTX PRO 6000 Blackwell: VRAM, speed and fit

GLM 4.5 Air at 8-bit runs on the 4× RTX PRO 6000 Blackwell with 128K of context at about 65 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× RTX PRO 6000 Blackwell · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
135 GB
of 384 GB
Context
128K
auto
Speed
65 tok/s
28 when full
First reply
0.2 s
2K prompt

Where the memory goes

Model
109 GB
Context
23.0 GB
Overhead
2.8 GB
OS
0.0 GB
Free
249 GB
Memory
135 GB of 384 GB
Context
128K tokens · auto
Speed
65 tok/s · 28 when full
First reply
0.2 s for a 2K prompt
Running cost
$33/mo at home · $287/mo on a cloud 4× A40

Validation: Validated

What GLM 4.5 Air on 4× RTX PRO 6000 Blackwell was checked against
CheckTargetResultStatus
Fits in GPU memory384 GB usable135 GB neededPass
Headroom5% free249 GB free (65%)Pass
Context for chat16K or more128KPass
Writing speed15 tok/s or more65 tok/sPass
First replyunder 3 s, 2K prompt0.2 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
GLM 4.5 Air
Parameters
110B, 15B active
Precision
8-bit (Q8_0)
Context
128K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
4× RTX PRO 6000 Blackwell
GPU can use
384 GB of 384 GB

Running cost

At home
$33/mo
1520 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 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 4× RTX PRO 6000 Blackwell run?

Other models that run on 4× RTX PRO 6000 Blackwell

Questions

Can GLM 4.5 Air run on 4× RTX PRO 6000 Blackwell?

GLM 4.5 Air at 8-bit runs on the 4× RTX PRO 6000 Blackwell with 128K of context at about 65 tokens a second.

Report VR-02QA7KB, calculated Nov 22, 2026 from published specs, not benchmarked.