Source URL: https://simonwillison.net/2025/Apr/5/llama-4/#atom-everything
Source: Simon Willison’s Weblog
Title: Quoting Ahmed Al-Dahle
Feedly Summary: The Llama series have been re-designed to use state of the art mixture-of-experts (MoE) architecture and natively trained with multimodality. We’re dropping Llama 4 Scout & Llama 4 Maverick, and previewing Llama 4 Behemoth.
Llama 4 Scout is highest performing small model with 17B activated parameters with 16 experts. It’s crazy fast, natively multimodal, and very smart. It achieves an industry leading 10M+ token context window and can also run on a single GPU!
Llama 4 Maverick is the best multimodal model in its class, beating GPT-4o and Gemini 2.0 Flash across a broad range of widely reported benchmarks, while achieving comparable results to the new DeepSeek v3 on reasoning and coding – at less than half the active parameters. It offers a best-in-class performance to cost ratio with an experimental chat version scoring ELO of 1417 on LMArena. It can also run on a single host!
Previewing Llama 4 Behemoth, our most powerful model yet and among the world’s smartest LLMs. Llama 4 Behemoth outperforms GPT4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks. Llama 4 Behemoth is still training, and we’re excited to share more details about it even while it’s still in flight.
— Ahmed Al-Dahle, VP and Head of GenAI at Meta
Tags: meta, llm-release, generative-ai, llama, ai, llms
AI Summary and Description: Yes
Summary: The text discusses the latest developments in the Llama series of large language models (LLMs) from Meta, highlighting their innovative architecture and performance capabilities. These advancements are particularly relevant for professionals in AI security, cloud computing, and infrastructure, as they have implications for model deployment and security.
Detailed Description: The text outlines the features and performance of the new models in the Llama series, specifically Llama 4 Scout, Llama 4 Maverick, and the upcoming Llama 4 Behemoth. Here are the major points of interest:
– **Llama 4 Scout**:
– Features 17 billion activated parameters with 16 experts.
– Notable for being the highest-performing small model in its category, characterized by rapid processing and a natively multimodal design.
– Can handle a significant context window of over 10 million tokens and is capable of running on a single GPU, emphasizing efficiency and accessibility for various applications.
– **Llama 4 Maverick**:
– Marketed as the best multimodal model, outperforming competitors such as GPT-4o and Gemini 2.0 Flash.
– Demonstrates superior reasoning and coding capabilities compared to other models while using fewer active parameters.
– Affords an excellent performance-to-cost ratio, with a notable ELO score of 1417 on LMArena, which suggests strong capabilities in competitive settings and real-world applications.
– Also designed to run on a single host, facilitating easier deployment.
– **Llama 4 Behemoth**:
– Touted as Meta’s most advanced model and among the leading LLMs globally.
– Expected to surpass the performance of models like GPT4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro across various STEM benchmarks.
– Currently in training, indicating ongoing advancements and potential future capabilities that could impact many sectors, including education and technical fields.
Overall, these developments present new opportunities and challenges in the realm of AI, particularly in ensuring security and compliance when deploying such powerful models in sensitive environments. The implications of using advanced LLMs like these necessitate robust security measures to prevent misuse and to protect data integrity, especially in cloud and infrastructure contexts.