Mark Zuckerberg led Meta has launched its latest AI models in the Llama 4 series that will power its chatbot by the same name on WhatsApp, Instagram and other company services. The two new Llama 4 models: Llama 4 Scout and Llama 4 Maverick are available to download from today from Meta website and Hugging Face.
Apart from these models, Meta also previewed another model called Llama 4 Behemoth which it says is “one of the smartest LLMs in the world and our most powerful yet to serve as a teacher for our new models.”
Meta has trained its Llama 4 models to be natively multimodal by pre-training them on vast amounts of unlabeled text, image, and video data. In simple terms, this means the models can understand and respond to both images and text seamlessly.
The new models take inspiration from Chinese AI startup DeepSeek by employing a new machine learning technique called mixture of experts which allows different parts of the model to be trained for specific tasks for better performance and efficiency.
However, none of the new models announced by Meta are reasoning models like OpenAI o3-mini or DeepSeek R1. Reasoning models take a bit more time to respond to answers and are aimed to mimicking human-like thinking to get better answers to more complex questions.
What you should know about new Llama 4 models:
Llama 4 Maverick contains 17 billion active parameters with 128 experts. Meta says Maverick is a “product workhorse model for general assistant and chat use cases” and is is ideal for precise image understanding and creative writing related tasks.
Meanwhile, Llama 4 Scout comes with 17 billion active parameters, 16 experts, and 109 billion total parameters. The smaller Llama 4 model is said to excel in tasks like document summarization and reasoning over code bases.
Scout comes with a context window of 10 million and is claimed to deliver better results across benchmarks from Gemma 3, Gemini 2.0 Flash Lite and Mistral 3.1.
How to use new Llama 4 models?
Llama 4 models has rolled out with Meta AI on WhatsApp, Instagram, Messenger and other company apps. The new features can also be used by going to Meta AI’s dedicated website in over 40 countries.
However, the multimodal features of Meta AI are currently limited to English users in the US, meaning no Ghibli style image generation with Meta AI just yet.
