Together AI Unveils Llama 3.2 APIs and Llama Stack for Enhanced Multimodal AI Development

Together AI has announced the launch of Llama 3.2 vision and lightweight models, alongside the release of Llama Stack, marking a significant milestone in open source AI development. This launch aims to streamline the creation of multimodal agentic applications, according to together.ai.

Key Offerings in Llama 3.2

The newly introduced Llama 3.2 suite includes:

• Free Llama 3.2 11B Vision Model: A no-cost option for developers to experiment with multimodal AI.
• Vision Models: Available in 11B and 90B parameters, optimized for tasks such as image captioning and visual question answering.
• Lightweight Model: A 3B model designed for faster inference with reduced resource consumption, ideal for cost-effective applications.

Llama Stack Integration

Together AI is among the first API providers for Llama Stack, which standardizes components necessary for building agentic, retrieval-augmented generation (RAG), and conversational applications. This integration aims to accelerate AI development by providing a robust framework.

Enterprise and Developer Applications

In collaboration with Meta, Together AI’s Llama 3.2 models are set to revolutionize various industries:

• Healthcare: Enhancing medical image analysis for better diagnostic accuracy.
• Retail & E-Commerce: Improving user experience with image and text-based searches.
• Finance & Legal: Streamlining workflows by analyzing graphical and textual content.
• Education & Training: Creating interactive tools that process both text and visuals.

Showcase: Napkins.dev

As part of the launch, Together AI is showcasing Napkins.dev, an open-source demo app that uses Llama 3.2 vision models to generate code from sketches, wireframes, or screenshots. This tool demonstrates the practical applications of Llama 3.2 in transforming conceptual designs into functional code.

Why Llama 3.2?

Llama 3.2 models support a range of features including a 128K context length and multi-language support, making them versatile for both multimodal image and text processing. The models are available in various configurations to meet different performance and cost requirements.

Getting Started

Developers and enterprises can explore Llama 3.2 on Together AI’s playground or integrate the models using the provided Python SDK. This flexibility aims to make it easier to build, fine-tune, and scale multimodal AI applications.

For more information, visit Together AI's official blog.