Skip to content

GLM-4.7 Flash on M3 Ultra 96 GB: VRAM, speed and fit

GLM-4.7 Flash at 8-bit runs on the M3 Ultra 96 GB with 198K of context at about 80 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
65.7 GB
of 96.0 GB
Context
198K
auto
Speed
80 tok/s
30 when full
First reply
1.6 s
2K prompt

Where the memory goes

Model
30.9 GB
Context
10.2 GB
Overhead
623 MB
OS
24.0 GB
Free
30.3 GB
Memory
65.7 GB of 96.0 GB
Context
198K tokens · auto
Speed
80 tok/s · 30 when full
First reply
1.6 s for a 2K prompt
Running cost
$3.55/mo at home · $66/mo on a cloud 2× RTX A5000

Validation: Validated

What GLM-4.7 Flash on M3 Ultra 96 GB was checked against
CheckTargetResultStatus
Fits in GPU memory72.0 GB usable41.7 GB neededPass
Headroom5% free30.3 GB free (42%)Pass
Context for chat16K or more198KPass
Writing speed15 tok/s or more80 tok/sPass
First replyunder 3 s, 2K prompt1.6 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-4.7 Flash
Parameters
31.2B, 3B active
Precision
8-bit (Q8_0)
Context
198K, 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 22, 2026.

How to run it

Ollama
ollama run glm-4.7-flash:latest
llama.cpp
llama-server -hf unsloth/GLM-4.7-Flash-GGUF
vLLM
vllm serve zai-org/GLM-4.7-Flash
MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-4.7-Flash-4bit

GLM-4.7 Flash on similar machines

All machines: GLM-4.7 Flash 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-4.7 Flash run on M3 Ultra 96 GB?

GLM-4.7 Flash at 8-bit runs on the M3 Ultra 96 GB with 198K of context at about 80 tokens a second.

Report VR-0GQHK3D, calculated Nov 22, 2026 from published specs, not benchmarked.