Introduction to Llama 4 Models

Meta has officially launched the initial two models from its Llama 4 suite, namely Llama 4 Scout and Llama 4 Maverick. According to the company, Maverick serves as the primary model, excelling in image and text understanding, particularly for general assistant and chat use cases, as detailed in a blog post. On the other hand, the smaller model, Scout, is capable of handling complex tasks such as multi-document summarization, parsing extensive user activity for personalized tasks, and reasoning over vast codebases. Additionally, Meta has introduced Llama 4 Behemoth, an upcoming model described as "among the world’s smartest LLMs," with CEO Mark Zuckerberg announcing that a fourth model, Llama 4 Reasoning, will be unveiled in the next month.

Availability of Llama 4 Models

Both Maverick and Scout are currently available for download from the Llama website and Hugging Face, and have been integrated into Meta AI, including WhatsApp, Messenger, and Instagram DMs.

Model Specifications

The Llama 4 models boast impressive specifications, with Scout featuring 17 billion active parameters and 16 experts. According to Zuckerberg, Scout is "extremely fast, natively multimodal, and has an industry-leading, nearly infinite 10 million token context length, and it is designed to run on a single GPU." In contrast, Maverick has 17 billion active parameters with 128 experts, outperforming competitors like GPT-4o and Gemini 2.0 on coding, reasoning, multilingual, long-context, and image benchmarks.

Upcoming Models

Zuckerberg has described the upcoming Behemoth model, which is still in training, as "the highest performing base model in the world," with 288 billion active parameters. Although it has not been released yet, it is likely that we will hear more about it and the Reasoning model soon, especially with Meta’s big AI developer conference, LlamaCon, taking place in a few weeks.

Conclusion

The release of the Llama 4 models marks a significant milestone in the development of AI technology, with Meta pushing the boundaries of what is possible with multimodal intelligence. As the company continues to innovate and improve its models, we can expect to see even more exciting advancements in the field of AI.

Note: The provided script has been removed as it was not part of the content to be rewritten.

Also, an image was present in the original text with a caption from Meta, but since it’s an image and not a text description, it cannot be directly rewritten. However, the information about the image is retained in the rewritten text.