Llama 4 Scout

Overview

Llama 4 Scout is Meta’s efficient open multimodal model: 109B MoE with 17B active parameters and an industry-first 10M token context window, small enough to run on a single H100 GPU.

Key features

• 10M token context window
• 109B MoE, 17B active
• Fits on one H100
• Native multimodal input
• Llama community license

Tags