Meta’s Llama 4 models now on relaxAI

Here at Civo, we’re proud to announce that we have become the first UK company to successfully host and operationalize Meta’s new Llama 4 model family on relaxAI, our AI assistant. This breakthrough positions relaxAI at the forefront of sovereign AI development, combining world-class capabilities with uncompromising data protection standards that UK businesses can trust.

According to Meta's official blog post, Llama 4 represents a significant step forward in AI capabilities, with a focus on multimodal intelligence. The release includes multiple models, including Scout and Maverick, which have been trained on a wide range of data.

With Llama 4, relaxAI customers can implement cutting-edge AI solutions while keeping their data within the UK, backed by Civo's UK sovereign guarantee. Our privacy-first framework is specifically designed to meet UK data protection requirements, and our close collaboration with Meta has enabled us to optimize Llama 4 performance on UK-hosted infrastructure, ensuring full GDPR compliance.

What is new with Llama 4?

The latest release of Llama models, Llama 4, introduces two new multimodal models: Llama 4 Scout and Llama 4 Maverick. The Llama 4 models are implemented with Meta's best practices and accelerated with the latest Nvidia GPUs, ensuring optimal performance. The models are also trained on a wide variety of images and video frames, enabling native multi-modal understanding. Llama 4 also features multilingual support, allowing users to have conversations in 12 different languages, with training data covering over 200 languages.

While not only representing a step forward in AI capabilities, it also gives users of relaxAI access to:

Scout (109B parameters)

Llama 4 Scout is a cutting-edge multimodal model that sets a new standard for performance and versatility. This model delivers exceptional performance with native multimodality and an industry-leading context length of 10 million. This vast context enables a wide range of applications, including multi-document summarization and reasoning over large codebases. With its state-of-the-art capabilities, Llama 4 Scout is poised to revolutionize various industries and use cases, from document analysis to AI-assisted coding.

Maverick (400B parameters)

Llama 4 Maverick is a highly advanced multimodal model that pushes the boundaries of AI capabilities. With 17 billion active parameters and 128 experts, totaling 400 billion parameters, Maverick achieves benchmark-topping results comparable to leading models like ChatGPT and Claude, making it ideal for complex AI problem-solving.

We've also enabled an agentic tool to call for this model, further expanding its capabilities and allowing it to tackle even more challenging tasks.

Mixture-of-Experts (MoE)

One of the key architectural breakthroughs in Llama 4 is the mixture-of-experts (MoE) architecture. This paradigm shift from traditional dense models enables dynamic neural pathways specialized for specific tasks, requiring 7.5x fewer calculations than a dense model of comparable size. As a result, Llama 4 achieves the same model performance at reduced computation demand, leading to increased latency and throughput.

What's coming next for Llama 4?

Meta has committed to releasing multiple iterations of the Llama 4 models throughout 2025. A key update includes:

Enhanced reasoning capabilities

A key focus on solving complex problems, building upon the systematic reasoning capabilities of the current Llama 4 models. With further reasoning capabilities, Llama 4 is expected to be able to tackle more nuanced and multi-step problems, providing more accurate and relevant responses.

Having further enhanced reasoning capabilities will allow for advancements in other areas such as multi-step problem solving, contextual understanding, and analogical reasoning. This upcoming enhancement will further solidify Llama 4's position as a cutting-edge AI solution.

Join the sovereign AI revolution

Join the sovereign AI revolution and take advantage of the cutting-edge AI capabilities offered by relaxAI. We invite you to experience the power of Llama 4 for yourself with the following options:

By choosing relaxAI, you can trust that your data will remain within the UK, backed by Civo's UK sovereign guarantee.

For more information on Llama 4 and relaxAI, check out our related articles:

• DeepSeek vs. Llama vs. GPT-4 AI Models
• Meet the New Llama 3.3