Llama 4 — local AI model by Meta
Written by Jakub Rusinowski · Last updated
Meta's groundbreaking Mixture-of-Experts (MoE) series. Llama 4 uses a sparse MoE architecture where only a fraction of parameters activate per token, delivering frontier-class intelligence with far lower hardware requirements than the total parameter count suggests.
Variants
The smallest Llama 4 variant needs about 67 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Llama 4 Scout 17B → 17B active / 109B total | ~66.6 GB | 10,000,000 | ollama run llama4:scout | |
| Llama 4 Maverick 17B → 17B active / 400B total | ~242.3 GB | 1,000,000 | ollama run llama4:maverick |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Llama 4 locally
Install Ollama, then pull the tag.
ollama run llama4:scoutPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
Applies to: Llama 4 Scout 17B, Llama 4 Maverick 17BRecommended GPU
The cheapest catalogued GPU that runs Llama 4 locally (min 67 GB VRAM) is the NVIDIA DGX Spark (128 GB).
Can I run Llama 4 on my GPU?
- Llama 4 on AMD Ryzen AI Max+ 395
- Llama 4 on Apple M1 Max
- Llama 4 on Apple M1 Pro
- Llama 4 on Apple M1 Ultra
- Llama 4 on Apple M2 Max
- Llama 4 on Apple M2 Pro
- Llama 4 on Apple M3 Max
- Llama 4 on Apple M3 Pro
- Llama 4 on Apple M3 Ultra
- Llama 4 on Apple M4
- Llama 4 on Apple M4 Max
- Llama 4 on Apple M5
- Llama 4 on Apple M5 Max
- Llama 4 on Apple M5 Pro
- Llama 4 on NVIDIA A100 80GB (PCIe)
- Llama 4 on NVIDIA DGX Spark
- Llama 4 on NVIDIA GeForce RTX 4090
- Llama 4 on NVIDIA GeForce RTX 5090
- Llama 4 on NVIDIA H100 80GB (PCIe)
- Llama 4 on NVIDIA L40S
- Llama 4 on NVIDIA RTX 6000 Ada Generation
- Llama 4 on NVIDIA RTX PRO 6000 Blackwell
Llama 4 — frequently asked questions
How much VRAM does Llama 4 need?
Llama 4 needs about 67 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Llama 4 Scout 17B (67 GB, Q4_K_M); Llama 4 Maverick 17B (242 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Llama 4 on an RTX 4090 (24 GB)?
Llama 4's smallest variant needs about 67 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Llama 4?
Q4_K_M is the best balance of quality and VRAM for Llama 4 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Llama 4 with Ollama?
Install Ollama, then run: ollama run llama4:scout. This downloads Llama 4 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.