On April 5, Meta unveiled its Llama 4 AI model series, which includes Llama 4 Maverick, designed for conversational applications, and Llama 4 Scout, optimized for processing large files. Utilizing a mixture of experts architecture allows the models to activate only relevant parts of the neural network for specific tasks, enhancing efficiency. Maverick features 128 experts and is noted for its competitive performance-to-cost ratio in conversations. Scout, boasting a remarkable 10 million token context window, is claimed to be the best multimodal model in its class.
Meta’s Llama 4 series introduces advanced AI models that leverage a mixture of experts architecture, enabling tailored solutions for conversation and large file processing.
Llama 4 Maverick emerges as a versatile model with 128 experts, boasting 17 billion active parameters, aimed at enhancing image interpretation and creative writing.
Despite Llama 4 Maverick's advertised performance efficiency, experts note discrepancies between its reported capabilities and the version shown to rank higher on LMArena.
Llama 4 Scout, with 17 billion active parameters and a context window of 10 million tokens, distinguishes itself as the leading multimodal model in its class.
Collection
[
|
...
]