Skip to content

Devstral Small 2 24B on Mac mini M4 Pro 48 GB: VRAM, speed and fit

Devstral Small 2 24B at FP8 runs on the M4 Pro 48 GB with 60K of context at about 8.7 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M4 Pro 48 GB · FP8 · Ollama / LM Studio · Chat

GPU memory
46.1 GB
of 48.0 GB
Context
60K
auto
Speed
8.7 tok/s
6.2 when full
First reply
16 s
2K prompt

Where the memory goes

Model
24.0 GB
Context
9.4 GB
Overhead
672 MB
OS
12.0 GB
Free
1.9 GB
Memory
46.1 GB of 48.0 GB
Context
60K tokens · auto
Speed
8.7 tok/s · 6.2 when full
First reply
16 s for a 2K prompt
Running cost
$1.84/mo at home · $61/mo on a cloud 2× RTX A4500

Validation: Not validated

What Devstral Small 2 24B on Mac mini M4 Pro 48 GB was checked against
CheckTargetResultStatus
Fits in GPU memory36.0 GB usable34.1 GB neededPass
Headroom5% free1.9 GB free (5%)Pass
Context for chat16K or more60KPass
Writing speed15 tok/s or more8.7 tok/sCheck
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
Devstral Small 2 24B
Parameters
24B
Precision
FP8
Context
60K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M4 Pro 48 GB
GPU can use
36.0 GB of 48.0 GB

Worth trying

  • 8-bit cache: Room for 112K context

Running cost

At home
$1.84/mo
84 W at the wall
Renting instead
$61/mo
2× RTX A4500 at $0.50/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 devstral-small-2:24b
llama.cpp
llama-server -hf unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF
vLLM
vllm serve mistralai/Devstral-Small-2-24B-Instruct-2512
MLX (Mac)
mlx_lm.chat --model mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit

Devstral Small 2 24B on similar machines

Other models that run on Mac mini M4 Pro 48 GB

Questions

Can Devstral Small 2 24B run on Mac mini M4 Pro 48 GB?

Devstral Small 2 24B at FP8 runs on the M4 Pro 48 GB with 60K of context at about 8.7 tokens a second.

Report VR-055U3WE, calculated Nov 22, 2026 from published specs, not benchmarked.