Monday, April 7, 2025

Llama 4 by Meta now live on Amazon Web Services

Related stories

Philips Wins F&S Award for CT 3500 Imaging Solution

Philips CT 3500, a scalable, adaptable, and durable imaging...

SandboxAQ Closes Series E with More Investors

New investments to accelerate SandboxAQ's innovation in quantitative AI...

Solo.io Unveils MCP Gateway for AI in Kubernetes

Open source MCP gateway provides advanced security, tool federation,...

Circus SE Launches CA-1 S4 AI Robot for Mass Market

The CA-1 Series 4 system features major upgrades across...

N-able Enhances Cyber Resiliency with Built-in Vulnerability Management

N-able UEM solutions now deliver a single, unified view...
spot_imgspot_img

Llama 4 can deliver more powerful results while using fewer computing resources—making advanced AI more accessible and cost-effective.

Amazon Web Services (AWS) has announced the availability of Meta’s latest Llama 4 models—Llama 4 Scout 17B and Llama 4 Maverick 17B—via Amazon SageMaker JumpStart. These advanced models will soon be accessible as fully managed, serverless offerings through Amazon Bedrock, further strengthening AWS’s growing portfolio of powerful generative AI solutions.

Both models come equipped with native multimodal capabilities, enabling them to understand and generate responses from both text and image prompts. In addition, they offer industry-leading context windows, allowing for significantly larger input processing compared to previous model versions. These enhancements make the Llama 4 models ideal for developing high-performance, efficient AI applications.

Expanding Choice and Innovation for Generative AI Developers

The introduction of Llama 4 Scout and Maverick adds to AWS’s already expansive selection of foundation models, reinforcing its mission to provide customers with the tools they need to build, customize, and scale next-gen AI solutions. AWS remains committed to bringing the latest innovations from top AI providers like Meta to its users the moment they are released, backed by enterprise-grade security and developer tools.

Why This Matters

This launch underscores AWS’s continued dedication to offering diverse and cutting-edge AI models. Llama 4 Scout 17B redefines scalability by supporting up to 10 million tokens in its context window—an almost 80x increase over its predecessor, Llama 3. This breakthrough enables more comprehensive tasks such as summarizing large volumes of documents, analyzing extensive user activity, or reasoning across entire codebases in a single go.

Meanwhile, Llama 4 Maverick 17B is built for versatility, excelling in both image and text comprehension across 12 languages, making it an ideal choice for multilingual chatbots, intelligent assistants, and global-facing applications.

Also Read: CaminoSoft Launches HorizonAX: A Next-Gen AI-Optimized Data Management Platform

Designed for Multimodality, Built for Efficiency

What sets the Llama 4 models apart is their native multimodal design, meaning they’re architected from the ground up to handle images and text as a unified input—rather than treating them separately.

At the core of both models is Meta’s new mixture of experts (MoE) architecture, which selectively activates only the most relevant parts of the model for each task. This results in significant computational efficiency, enabling organizations to achieve higher performance at lower operational costs—a major win for both innovation and sustainability in AI deployment.

Get to Know the Models

If the Llama 4 models were people, Scout would be the detail-obsessed research assistant, capable of instantly retrieving insights from massive datasets and anticipating informational needs with uncanny accuracy. Maverick, on the other hand, would be the multilingual creative director, skilled at visual storytelling, language precision, and brand consistency across diverse platforms.

Under the Hood: Key Specs

  • Llama 4 Scout 17B includes 17 billion active parameters and 109 billion total parameters, enabling state-of-the-art performance in its class. Its standout feature—a 10 million token context window—allows it to process far more data in one pass than any of its predecessors.

  • Llama 4 Maverick 17B also boasts 17 billion active parameters, but with 400 billion total parameters distributed across 128 expert modules. Its MoE architecture ensures only the most relevant experts activate per task, combining power with efficiency.

Smarter Design, Better Results

The MoE framework essentially mirrors a real-world team of specialists: instead of one generalist model tackling all tasks, it routes requests to the most qualified components. This method results in smarter processing, faster insights, and more efficient compute usage—making the technology accessible to a wider range of businesses and developers.

From a developer’s perspective, these models unlock the ability to build highly intelligent applications that can manage complex tasks like multimodal understanding and multilingual communication—at scale.

What’s Next?

AWS continues to prioritize rapid availability of cutting-edge AI tools. As more Llama 4 variants are released, developers can expect broader access to models tailored across different sizes and modalities, helping them harness the full potential of generative AI—securely and at scale.

Subscribe

- Never miss a story with notifications


    Latest stories

    spot_img