Part 5/9:
Llama 4 Maverick
Targeted as a more accessible version, Maverick incorporates 17 billion active parameters and boasts 128 experts. It is designed for more general use, offering 1 million context tokens, thus facilitating complex tasks while maintaining a great performance-cost ratio.
Llama 4 Scout
Positioned as the most user-friendly version, Scout features 17 billion parameters and a massive 10 million context window. This unique feature is expected to revolutionize tokenization by allowing a more straightforward approach to managing language inputs.