Llama 3 Models — Hardware Requirements

65 Llama 3 models from Meta and the community, from the smallest that runs in 1.6 GB of VRAM at Q4 up to 405.9B parameters. Every row links to full quantization tables and GPU compatibility.

All Llama 3 Models by Size

How Llama 3 Compares — llmrun Composite Score

Llama 3.1 405B Instruct is the highest-scoring Llama 3 model with an llmrun Composite Score of 83.5 — #82 among 106 open models. The top proprietary model, Gemini 4 Carbon, scores 158.6. Click a model to see its full benchmark breakdown.

Frequently Asked Questions

How much VRAM do I need to run a Llama 3 model?

The smallest Llama 3 model, Llama 3.2 1B MathCodeInstruct 5k, runs from 1.6 GB of VRAM at Q4_K_M (smaller low-bit quantizations start at 0.9 GB with a noticeable quality loss). Larger family members need proportionally more — see the table above for every model.

Which Llama 3 models can I run on a 16 GB GPU?

38 of 60 Llama 3 models fit in 16 GB of VRAM at Q4_K_M, including Llama 3.2 1B Instruct, Llama 3.1 8B Instruct, Llama 3.2 3B Instruct.

What is the most popular Llama 3 model to run locally?

Llama 3.2 1B Instruct is the most downloaded Llama 3 model in local-friendly quantized formats. It runs from 2.3 GB of VRAM at Q4_K_M.

How do Llama 3 models score on benchmarks?

Llama 3.1 405B Instruct leads the family with an llmrun Composite Score of 83.5, ranking #82 among 106 open models, while the top proprietary model, Gemini 4 Carbon, scores 158.6. See the comparison chart above for the full standings.