Skip to content

LFM2.5 24B A2B on M3 Ultra 96 GB: VRAM, speed and fit

LFM2.5 24B A2B at 8-bit runs on the M3 Ultra 96 GB with 128K of context at about 104 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
50.8 GB
of 96.0 GB
Context
128K
auto
Speed
104 tok/s
68 when full
First reply
1.2 s
2K prompt

Where the memory goes

Model
23.7 GB
Context
2.5 GB
Overhead
570 MB
OS
24.0 GB
Free
45.2 GB
Memory
50.8 GB of 96.0 GB
Context
128K tokens · auto
Speed
104 tok/s · 68 when full
First reply
1.2 s for a 2K prompt
Running cost
$3.55/mo at home · $58/mo on a cloud 2× RTX 2000 Ada

Validation: Validated

What LFM2.5 24B A2B on M3 Ultra 96 GB was checked against
CheckTargetResultStatus
Fits in GPU memory72.0 GB usable26.8 GB neededPass
Headroom5% free45.2 GB free (63%)Pass
Context for chat16K or more128KPass
Writing speed15 tok/s or more104 tok/sPass
First replyunder 3 s, 2K prompt1.2 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
LFM2.5 24B A2B
Parameters
24B, 2B active
Precision
8-bit (Q8_0)
Context
128K, 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
$58/mo
2× RTX 2000 Ada at $0.48/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 lfm2.5:24b
llama.cpp
llama-server -hf unsloth/LFM2.5-24B-A2B-GGUF
vLLM
vllm serve LiquidAI/LFM2.5-24B-A2B
MLX (Mac)
mlx_lm.chat --model mlx-community/LFM2.5-24B-A2B-OptiQ-4bit

LFM2.5 24B A2B on similar machines

All machines: LFM2.5 24B A2B 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 LFM2.5 24B A2B run on M3 Ultra 96 GB?

LFM2.5 24B A2B at 8-bit runs on the M3 Ultra 96 GB with 128K of context at about 104 tokens a second.

Report VR-0ZKV0BN, calculated Nov 22, 2026 from published specs, not benchmarked.