It was recently announced that it will be held on April 4th, US West Coast timeThe first LlamaCon developer eventLater, Meta announced the release of two versions of the Llama 4 large-scale natural language model:LLama 4 Scout and Llama 4 Maverick.
Among them, Llama 4 Maverick will target mainstream applications and can be used for digital assistant services and chatbots, while the smaller LLama 4 Scout can process document content summaries, analyze large amounts of user activity, and provide personalized interactions or make inferences about large-scale program code content.
As for parameter specifications, LLama 4 Scout has 170 billion sets of parameters and corresponds to 16 professional field inferences. It also supports multimodal operations and executes very quickly. It can provide contextual inferences for up to 1000 million words. It is designed to run on a single GPU and emphasizes flexible deployment on various network architectures.
Llama 4 Maverick also has 170 billion parameters and corresponds to 128 professional field inferences. It surpasses OpenAI's GPT-4o and Google's Gemini 2.0 in projects such as encoding, inference, multilingual understanding, longer context analysis, and image benchmarking. Its inference and encoding processing performance is even similar to DeepSeek v3.1.
Meanwhile, Meta also announced the launch of the larger Llama 4 Behemoth, featuring 2880 billion parameters, aiming to compete with other larger AI models. Meta will also announce LLama 5 Reasoning in May, which will improve inference performance. Details are expected to be revealed at the inaugural LlamaCon developer event.
Currently, LLama 4 Scout and Llama 4 Maverick are available through the LLama official website and Hugging Face hosting website, and are integrated into Meta AI application resources, and are applied to Meta's WhatsApp, Messenger, and Instagram DM messaging functions.
