Text model · LFM

LF LFM2.5 8B-A1B: RAM and VRAM requirements

LFM2.5 8B-A1B needs about 6.7 GB to run at Q4_K_M (the Q4_K_M GGUF file is ~5.2 GB to download; KV cache and overhead add the rest), or about 10.5 GB at Q8_0. The lightest hardware that runs it is Nvidia GeForce RTX 3060 (12GB).

LFM family · 8.3B params (Mixture-of-Experts: activates only 1.5B of 8.3B params per token, so generation is faster than the total size suggests) · released May 2026.

Shopping for hardware? See what runs LFM2.5 8B-A1B →

License LFM Open License v1.0 · Conditional ↓ 171.3K/mo ♥ 690 on HuggingFace

Q4_K_M GGUF

5.2 GB

Q8_0 GGUF

9 GB

Memory @ Q4 (4k)

~6.7 GB

Context

128 k

Q4_K_M and Q8_0 are GGUF download (file) sizes. Memory @ Q4 includes the KV cache and overhead, so it runs larger than the file on disk.

Will it run on your device?

LFM2.5 8B-A1B runs on 33 of 40 tracked devices at Q4_K_M.

33 run well 0 tight fit 7 too small

Fit check Q4_K_M

Your device

Apple M4 (16GB)

Yes, it runs fast

needs 6.7 GB usable 10.5 GB

See the full breakdown

$ollama run lfm2.5:8b-a1b

Memory breakdown

Weights (Q4_K_M)5.2 GB

KV cache (4k)0.7 GB

Overhead0.8 GB

Total6.7 GB

How context length changes it

4k context ~6.7 GB 32k context ~11.5 GB 128k context ~28.1 GB

Longer context grows the KV cache, which for LFM2.5 8B-A1B is sized by its 1.5B active params, not the full 8.3B. It needs ~6.7 GB at 4k and ~28.1 GB at 128k.

Quantization sizes

GGUF quantson disk

Quantization	Size on disk
Q2_K	3.5 GB est
Q3_K_M	4.1 GB est
Q4_K_M (default)	5.2 GB
Q5_K_M	5.9 GB est
Q6_K	6.8 GB est
Q8_0	9 GB
FP16	17 GB

Lower quant = smaller and faster, slightly lower quality. Q4_K_M is the common default.

Run it

Ollama

$ ollama run lfm2.5:8b-a1b

llama.cpp

$ llama-cli -hf LiquidAI/LFM2.5-8B-A1B-GGUF:Q4_K_M

LM Studio

$ lms get LiquidAI/LFM2.5-8B-A1B-GGUF

Which devices can run LFM2.5 8B-A1B?

Apple Silicon Macs

RAM-only laptops

iPhone & iPad

Android

NVIDIA GPUs

AMD GPUs

Same job, different size

Pick by what fits your memory: step down to free up VRAM, or step up if you have headroom.

Next size down Qwen2.5-VL 7B 8.29B Next size up Granite 4.1 8B 8.8B

Similar models

Head-to-head

FAQ

How much VRAM or RAM does LFM2.5 8B-A1B need?

At Q4_K_M, LFM2.5 8B-A1B needs about 6.7 GB (weights ~5.2 GB + KV cache + overhead) at a 4k context. At Q8_0 budget ~10.5 GB.

What is the Q4_K_M GGUF file size of LFM2.5 8B-A1B?

The Q4_K_M GGUF file is about 5.2 GB to download, and the Q8_0 GGUF is about 9 GB. That is the weights file on disk; to run it you also need room for the KV cache and overhead, so budget ~6.7 GB of memory at Q4_K_M.

Can LFM2.5 8B-A1B run on a laptop?

LFM2.5 8B-A1B is large; you need a 24 GB+ GPU or a 32-48 GB Mac at Q4_K_M.

Is LFM2.5 8B-A1B cheaper to run because it is a MoE model?

It is faster, not lighter. LFM2.5 8B-A1B activates only 1.5B of 8.3B params per token (so it runs quickly), but all experts must stay in memory, so it still needs memory for the full 8.3B.

Can I use LFM2.5 8B-A1B commercially?

Conditionally. LFM Open License: free under $10M revenue; larger companies need a commercial license.

Understand the numbers

Short guides to the ideas behind LFM2.5 8B-A1B's memory and quant figures.

Liquid AI's reasoning-tuned MoE for on-device use: 8.3B total, 1.5B active, 128K context. Despite the A1B name the vendor card states 1.5B active params. Sizes from the Ollama tag listing, matching the vendor GGUF repo.

Sources

Last validated 2026-08-03. Memory figures are estimates. See methodology.