Skip to content

Diffusiongemma 26B A4B IT on Mac mini M4 24 GB: VRAM, speed and fit

Diffusiongemma 26B A4B IT at 3-bit runs on the M4 24 GB with 120K of context at about 42 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M4 24 GB · 3-bit (Q3_K_M) · Ollama / LM Studio · Chat

GPU memory
23.1 GB
of 24.0 GB
Context
120K
auto
Speed
42 tok/s
19 when full
First reply
16 s
2K prompt

Where the memory goes

Model
11.8 GB
Context
2.5 GB
Overhead
856 MB
OS
8.0 GB
Free
891 MB
Memory
23.1 GB of 24.0 GB
Context
120K tokens · auto
Speed
42 tok/s · 19 when full
First reply
16 s for a 2K prompt
Running cost
$0.85/mo at home · $29/mo on a cloud RTX 2000 Ada

Validation: Not validated

What Diffusiongemma 26B A4B IT on Mac mini M4 24 GB was checked against
CheckTargetResultStatus
Fits in GPU memory16.0 GB usable15.1 GB neededPass
Headroom5% free891 MB free (5%)Pass
Context for chat16K or more120KPass
Writing speed15 tok/s or more42 tok/sPass
First replyunder 3 s, 2K prompt16 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
Diffusiongemma 26B A4B IT
Parameters
25.8B, 4B active
Precision
3-bit (Q3_K_M)
Context
120K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M4 24 GB
GPU can use
16.0 GB of 24.0 GB

Running cost

At home
$0.85/mo
39 W at the wall
Renting instead
$29/mo
RTX 2000 Ada at $0.24/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/diffusiongemma-26B-A4B-it-GGUF
vLLM
vllm serve google/diffusiongemma-26B-A4B-it
MLX (Mac)
mlx_lm.chat --model mlx-community/diffusiongemma-26B-A4B-it-4bit

Diffusiongemma 26B A4B IT on similar machines

Other models that run on Mac mini M4 24 GB

Questions

Can Diffusiongemma 26B A4B IT run on Mac mini M4 24 GB?

Diffusiongemma 26B A4B IT at 3-bit runs on the M4 24 GB with 120K of context at about 42 tokens a second.

Report VR-1N21GUD, calculated Nov 22, 2026 from published specs, not benchmarked.