Meta has officially launched its latest artificial intelligence models, Llama 4 Scout and Llama 4 Maverick marking a major step forward in the company’s AI development.
These are Meta’s first multimodal large language models, meaning they can process not just text but also images, videos, and even audio.
This opens up a wide range of possibilities for real-world applications across different industries.
Llama 4 Scout is built for efficiency.
It uses 17 billion active parameters and works with 16 AI “experts”, totaling 109 billion parameters.
What’s impressive is that it can run on just one high-end graphics card, like the NVIDIA H100—making it more accessible and energy-saving.
Scout also features a 10 million-token context window, allowing it to handle long texts or conversations without forgetting earlier parts.
For more advanced work, Meta introduced Llama 4 Maverick, which has 400 billion parameters across 128 experts. Despite its size, it’s optimized for performance and efficiency.
It supports up to 1 million tokens in context, and has outperformed other top AI models like GPT-4o and Gemini 2.0 Flash in tasks like reasoning, coding, and image analysis.
Meta also gave a sneak peek at a much larger model called Llama 4 Behemoth.
It’s meant to be one of the most advanced AI systems they’ve ever built and will help train future AI models.
With all of this, Meta is showing that it’s serious about AI.
Meta also plans to spend up to $65 billion this year to improve its AI tech, hoping to keep up with other big names in the field.
Multimodal means that these AI models can work with different types of data at the same time. For example, they can understand and respond to both text and images together, making them more powerful for tasks like answering questions about pictures or interpreting complex content.
Llama 4 Scout is designed for efficiency. It uses 17 billion active parameters and can run on just one high-end GPU (like the NVIDIA H100), making it energy-efficient and affordable. Despite its smaller size, it can handle long conversations and documents without losing context.
Llama 4 Maverick is more advanced, with 400 billion parameters and 128 experts. It’s designed to handle more complex tasks, like reasoning and image understanding, and can process up to 1 million tokens in one go. It has been shown to outperform other top AI models in benchmarks.
Keep Reading: Level Up Gaming: Nintendo Switch 2 Launches June 2025 with Exciting Features and Games
Leave a Reply