Llama 4 Scout 109B
By Meta · United States
Updated 2026-07-13
Overview
Meta's compact Llama 4 MoE — 109B total, 17B active, natively multimodal, with an unprecedented 10M token context. Fits on a single H100.
When to pick this model
- Whole-codebase or whole-corpus analysis up to 10M tokens
- Multimodal pipelines where one H100 is the inference budget
- Long-form document understanding without RAG
- Multilingual chat with native image input
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 65 GB |
| Q5_K_M | 78 GB |
| Q8_0 | 117 GB |
| FP16 (no quantization) | 218 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, Llama 4 Scout 109B is server-class even at Q4_K_M (65 GB). Stepping up to Q8_0 nearly doubles the footprint to 117 GB, and unquantized FP16 weights take 218 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, Llama 4 Scout 109B needs roughly 100 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 3 tokens/sec on entry-level GPUs, on the order of 12 tokens/sec on a mid-range card, and up to 30 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches Llama 4 Scout 109B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for Llama 4 Scout 109B |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 65 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 65 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Does not fit — needs 65 GB at Q4_K_M |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Does not fit — needs 65 GB at Q4_K_M |
| 32 GB | RTX 5090 | Does not fit — needs 65 GB at Q4_K_M |
Which GPU should you buy to run Llama 4 Scout 109B?
To run Llama 4 Scout 109B locally at Q4, you need ~65 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Published benchmark scores
| Benchmark | Score |
|---|---|
| MMLU-Pro | 74 |
Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.
Strengths
- 10M token context — unmatched among open models
- Runs on a single H100 thanks to MoE sparsity
- Native multimodal input — no separate vision adapter needed
- 17B active parameters keeps inference fast
Limitations
- Hugging Face gated access
- Llama 4 Community License with the >700M MAU clause
- Long-context quality drops well before the 10M ceiling
- Newer than Llama 3.1 — tooling still catching up
Typical workloads
In our catalog grid, Llama 4 Scout 109B is filed under Extreme Long Context, Vision, Agents — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: vision-language work — screenshots, charts, scanned documents; multilingual workloads.
The 9765k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. It ships under the Llama 4 Community license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: MoE 16 experts · 109B/17B active · iRoPE · natively multimodal
Training: Meta Llama 4 compact flagship.
The long-context champion of open weights — if you actually need 10M tokens, nothing else comes close on a single H100.
Quick start
ollama run llama4:scoutOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does Llama 4 Scout 109B need?
At the recommended Q4_K_M quantization, Llama 4 Scout 109B needs about 65 GB of VRAM. Q8_0 takes 117 GB, and unquantized FP16 weights take 218 GB.
Can Llama 4 Scout 109B run without a GPU?
Yes — with roughly 100 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does Llama 4 Scout 109B support?
Llama 4 Scout 109B supports a 9765k-token context window (10,000,000 tokens).
Can I use Llama 4 Scout 109B commercially?
Llama 4 Scout 109B ships under the Llama 4 Community license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is Llama 4 Scout 109B on consumer hardware?
Our compatibility engine estimates on the order of 12 tokens/sec on a mid-range GPU and up to 30 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of Llama 4 Scout 109B should I download first?
Start with Q4_K_M (65 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.