Llama logo

Llama 4 Scout

Llama 4 Scout loads in 67 GB at Q4_K_M and decodes at the speed of its 17B active path, comfortably from 96 GB of memory, Macs included, starting at the MacBook Pro M5 Max 128GB. Below is the memory math we ran for it, the smallest card that fits, and the fastest machine we track.

PARAMETERS
109B (17B active)
FORMAT
Q4_K_M
MIN MEMORY
96 GB
BEST FOR
Long context, Quality, Multimodal

How much memory does Llama 4 Scout need?

Llama 4 Scout is a 109B mixture-of-experts model (17B active per token) built for long context, quality, multimodal. The weights alone take 67 GB at Q4_K_M; add context and the OS, and it belongs in the 96 GB memory class. Every number below comes from the ModelFit engine: speeds are estimates, labeled est., because we do not benchmark hardware ourselves.

Weights (Q4_K_M)
67 GB
+ KV cache (16k)
~6.0 GB
Total at 16k
~73.0 GB
Comfortable from
96 GB

KV = fp16 estimate (q8_0 cache roughly halves it). "Comfortable" = weights + KV within the engine's tiered budget (~70-85% of memory).

Which machines run Llama 4 Scout?

Smallest GPU
~35 tok/s est. · 96 GB VRAM
Fastest
~35 tok/s est.
Runs on a Mac?
~16 tok/s est.

Verdicts come from the same engine that powers our fit checker: memory capacity first, bandwidth for speed. How we compute these numbers · Hugging Face model card

Go deeper

Every Llama 4 Scout quant by real GGUF file sizeBest hardware for Llama 4 Scout: smallest GPU, fastest, Macs

More Llama models

Llama 3.2 1B InstructLlama 3.2 3B InstructLlama 3.1 8B InstructLlama 3.1 70B InstructLlama 3.3 70B InstructLlama 4 MaverickLlama 3.1 405B Instruct

Frequently asked questions

How much memory does Llama 4 Scout need?

67 GB for the Q4_K_M weights, plus ~6.0 GB of KV-cache at 16k context, about 73.0 GB total. Comfortable from 96 GB of VRAM or unified memory.

Does Llama 4 Scout run on a Mac?

Yes, from the MacBook Pro M5 Max 128GB (~16 tok/s est.). Unified memory means the RAM budget is the only limit.

What is the smallest GPU for Llama 4 Scout?

The NVIDIA RTX PRO 6000 Blackwell (96 GB VRAM) is the smallest tracked card that runs Llama 4 Scout comfortably, ~35 tok/s est.

Which Llama 4 Scout quant should I pick?

The tracked Q4_K_M build at 67 GB is the one we recommend for most machines. Every published build, ranked by real GGUF file size rather than a formula, is on the quant comparison page linked below.

Cite this page

ModelFit: Llama 4 Scout, specs, memory math and hardware verdicts.
https://modelfit.io/models/llama4-scout/ (dataset updated 2026-10-13, CC BY 4.0).