Meta has unveiled Llama 4, its latest family of large language models (LLMs), as part of its intensified efforts to compete with leading players in the artificial intelligence space.
The new release, which includes a suite of multimodal models capable of processing text, images, audio, and video, is seen as a major step in the company’s ambition to shape the future of AI.
The three-tiered release includes Llama 4 Scout, Llama 4 Maverick, and the forthcoming Llama 4 Behemoth, each designed to handle increasingly complex tasks using Meta’s advanced “mixture-of-experts” architecture. This method enables models to activate only the most relevant parameters for each task, significantly improving efficiency and performance.
Speaking on the launch, Meta CEO Mark Zuckerberg described the models as a “significant leap” towards building general-purpose AI systems.
“We’ve engineered Llama 4 to be faster, smarter, and more capable of handling real-world challenges in both consumer and enterprise settings,” he said.
Also Read:Â Meta Deletes 100 Million Fake Facebook Pages in Major Anti-Spam Drive
Among its standout features is Llama 4 Scout, a lightweight model with 17 billion active parameters and a context window of up to 10 million tokens—enabling it to manage extended conversations with enhanced coherence. Designed for accessibility, it can run on a single Nvidia H100 GPU.
The Llama 4 Maverick model, with 400 billion parameters and 128 experts, targets more demanding applications, such as code generation, in-depth reasoning, and advanced decision-making. The flagship Behemoth, Â still in training,is expected to become one of the largest models globally with 2 trillion total parameters.
Meta has released the models under an “open-weight” framework, allowing researchers and developers to fine-tune and deploy the systems under specific licensing terms. While not fully open-source, the approach aims to balance innovation with responsible AI development.
In a move to accelerate adoption, Llama 4 models are now accessible via cloud platforms including Amazon Web Services (AWS) and Cloudflare Workers AI. This broad availability allows developers to integrate the models into various applications across industries.
The company also highlighted improvements in content moderation and ethical responsiveness. According to Meta, Llama 4 models are more likely to respond to socially or politically sensitive questions, declining less than 2% of such prompts, a reduction from previous iterations.
