Written by Jakub Rusinowski · Last updated April 5, 2025
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.
| Licence | What it permits | Applies to |
|---|---|---|
Llama Community | Commercial use permitted Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms. | Llama 4 Scout 17B, Llama 4 Maverick 17B |
| Llama 4 Scout 17B | Min 67 GB VRAM · Q4_K_M · 10,000,000 ctx · ollama run llama4:scout |
| Llama 4 Maverick 17B | Min 242 GB VRAM · Q4_K_M · 1,000,000 ctx · ollama run llama4:maverick |
The cheapest GPU that runs Llama 4 locally (min 67 GB VRAM) is the AMD Ryzen AI Max+ 395 (96 GB).
Install Ollama then run: ollama run llama4:scout
Minimum VRAM: 67 GB. For best results use Q4_K_M quantization.
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.
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.
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.
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.