Llama 4 Scout

Llama 4 Scout is a high-efficiency edge model optimized for real-time autonomous navigation and spatial reasoning.

Engineered for low-latency environments: Llama 4 Scout delivers 120 tokens per second on mobile hardware (Snapdragon 8 Gen 3) while maintaining a 128k context window. It excels at visual-spatial mapping and sensor fusion: processing raw LIDAR and camera feeds into actionable navigation commands. This 8B-parameter variant outperforms previous iterations by 40% in power efficiency: making it the standard for drone fleets and industrial robotics requiring on-device intelligence without cloud dependency.

https://ai.meta.com/llama/llama-4-scout