Meta has officially launched its latest AI models, Llama 4 Scout and Llama 4 Maverick, marking a significant advancement in multimodal artificial intelligence. Meta announced on its blog on Saturday. These models are designed to enhance personalized experiences across various applications, featuring groundbreaking capabilities that set them apart from previous generations.
Llama 4 Scout, with 17 billion active parameters and a context window of 10 million tokens, has been recognized as the best in its class. It outperforms earlier Llama models and competitors like Gemma 3, delivering superior results across widely reported benchmarks.
Llama 4 Maverick, also featuring 17 billion active parameters but with 128 experts, has demonstrated exceptional performance against notable models such as GPT-4o and Gemini 2.0., according to Meta. It achieves a remarkable performance-to-cost ratio, making it a compelling choice for developers looking to integrate advanced AI capabilities into their products.
These new models are built on a mixture-of-experts (MoE) architecture, which enhances efficiency by activating only a fraction of the total parameters during inference. This innovation allows Llama 4 Scout and Maverick to deliver high-quality results while maintaining lower operational costs.
Meta is also previewing Llama 4 Behemoth, a 288 billion active parameter model still in training, which is expected to surpass current leading models in STEM benchmarks. While not yet available for public use, further details will be shared as development progresses.
Both Llama 4 Scout and Maverick are now available for download on llama.com and Hugging Face, with integration options across Meta’s platforms, including WhatsApp and Instagram.
As artificial intelligence continues to play a crucial role in daily life, Meta emphasizes the importance of open access to these advanced models, enabling developers and enterprises to create innovative solutions that leverage the power of AI.