On April 5, 2025, Meta Platforms unveiled its latest advancements in artificial intelligence with the introduction of the Llama 4 series, which includes three distinct models: Llama 4 Scout, Llama 4 Maverick, and the yet-to-be-released Llama 4 Behemoth. These new models are designed to enhance the capabilities of AI by integrating various data types such as text, images, and audio, marking a significant step forward in the realm of multimodal AI systems.
The Llama 4 Scout is built to cater to users who require efficient processing of diverse information. It boasts 17 billion active parameters and utilizes a Combination of Experts (MoE) approach, trained with 16 specialists. This model can operate on a single Nvidia GPU H100, making it accessible for smaller businesses and professionals engaged in less complex tasks. According to Meta, the Scout model outperforms competitors like Google's Gemma 3 and Gemini 2.0 Flash-Lite, as well as Mistral's Mistral 3.1 in various benchmark tests.
In contrast, the Llama 4 Maverick, also featuring 17 billion parameters, was trained with 128 experts, allowing it to handle more demanding applications. This model is said to surpass OpenAI's GPT-4o and Google's Gemini 2.0 Flash across a wide range of benchmarks. Interestingly, while the Maverick matches the performance of DeepSeek v3 in areas such as reasoning and coding, it does so with fewer active parameters, illustrating Meta's commitment to efficiency and capability.
Both the Scout and Maverick models are now available for download on llama.com and Hugging Face, providing developers and researchers the opportunity to integrate these advanced tools into their own applications. Meta has emphasized its goal of fostering collaboration within the open-source community, enhancing the overall development of AI technologies.
Meanwhile, the Llama 4 Behemoth is still under development, boasting an impressive 288 billion parameters. Meta describes it as the most powerful model in the Llama 4 series, designed to excel in benchmarks focused on mathematics, science, and technology. Preliminary results indicate that the Behemoth outperforms GPT-4.5, Claude Sonnet 3.7 from Anthropic, and Gemini 2.0 Pro in these critical areas. However, it is important to note that the Behemoth is not yet available for public testing.
In recent months, Meta has faced challenges during the development of Llama 4, as reported by The Information. The company had to delay the launch due to the model not meeting internal expectations in certain technical criteria, particularly in reasoning and mathematical tasks. There were also concerns regarding its ability to conduct human-like voice conversations, which are increasingly important in AI interactions. Despite these hurdles, Meta is committed to investing up to $65 billion this year to expand its AI infrastructure, reflecting the competitive pressure to deliver advanced AI solutions.
Meta's Llama 4 series is part of a broader trend among tech giants to enhance their AI capabilities following the success of OpenAI's ChatGPT. This shift has led to increased investments in machine learning and AI technologies across the industry, as companies strive to stay ahead in a rapidly evolving landscape.
As the Llama 4 models roll out, users can begin testing the capabilities of Meta AI across platforms such as WhatsApp, Messenger, and Instagram Direct. The update is set to launch in over 40 countries and 13 languages, although the advanced multimodal features will initially be limited to English-speaking users in the United States.
Meta's push towards open-source software with the Scout and Maverick models is a strategic move to engage the developer community and encourage innovation. By allowing developers to customize and build upon these models, Meta aims to create a robust ecosystem that benefits from collective expertise and creativity.
In summary, the Llama 4 series represents a significant leap forward in Meta's AI capabilities, with the Scout and Maverick models already available for public use, while the Behemoth promises to set new standards in AI performance upon its eventual release. As Meta continues to refine its offerings, the implications for businesses and developers are profound, paving the way for more sophisticated AI applications in various fields.
