Introducing Amazon Nova: A New Generation of AI Foundation Models

Amazon has introduced Amazon Nova, a new generation of foundation models (FMs) that deliver advanced intelligence and creative capabilities across text, image, and video modalities. Integrated into Amazon Bedrock, a managed service providing access to leading FMs via a single API, Amazon Nova represents a significant leap in AI innovation. Building on custom-built Inferentia and Trainium chips and over 1,000 generative AI applications already in use, these models address challenges in latency, cost, customization, and agentic capabilities while offering transformative value to users.

Amazon Nova includes six models: Nova Micro, a text-only model optimized for speed and cost; Nova Lite, a multimodal model for processing text, images, and video; Nova Pro, combining accuracy, speed, and versatility; Nova Premier, tailored for complex reasoning and teaching custom models (launching Q1 2025); Nova Canvas, for studio-quality image generation; and Nova Reel, for high-quality video creation. Supporting 200 languages, these models process up to 300K tokens and are set to exceed 2M tokens by 2025. Amazon Nova Micro, Lite, and Pro are 75% cheaper and faster than competitors in their classes.

Amazon Nova models have achieved exceptional results in industry benchmarks. Nova Micro outperformed Meta's LLaMa 3.1 8B and Google Gemini 1.5 Flash-8B across benchmarks, while Nova Lite surpassed OpenAI's GPT-4o mini, Anthropic's Claude Haiku 3.5, and Google Gemini 1.5 Flash-8B on multimodal tasks. Nova Pro excelled in Retrieval Augmented Generation (RAG) and multimodal workflows, beating leading models in accuracy and speed. These results showcase Nova's ability to handle diverse tasks, from understanding complex visuals to executing multi step workflows.

Nova's creative capabilities empower users to generate professional-grade images and videos. Nova Canvas creates high-quality images with tools for safe AI practices like watermarking and content moderation, outperforming DALL-E 3 and Stable Diffusion. Nova Reel enables video creation using natural language prompts, with plans to expand from six-second clips to two-minute videos by mid-2025. These tools are being adopted by companies like Dentsu Digital Inc. and Musixmatch, revolutionizing creative processes and reducing production times.

Seamlessly integrated with Amazon Bedrock, Nova allows users to experiment, fine-tune models with proprietary data, and use distillation for efficient performance. RAG ensures responses are grounded in customer data, while agentic optimizations support complex tasks via API interactions. These capabilities enable users to streamline operations and enhance innovation across industries. Responsible AI practices are at the forefront, with integrated safety features and AWS AI Service Cards ensuring ethical use.

Strategic partners like SAP, Deloitte, and Palantir leverage Nova's advanced capabilities. SAP integrates Nova into its AI Core for personalized solutions, while Deloitte delivers generative AI services globally. Palantir employs Nova Pro to improve decision-making across industries. Other companies like Shutterstock and Caylent use Nova to enhance content creation and operational efficiency.

Plans include a speech-to-speech model launching in Q1 2025 for natural conversational AI and a multimodal-to-multimodal model arriving in mid-2025 for versatile content generation across text, images, audio, and video. These developments aim to simplify complex applications and expand Nova's transformative potential. Amazon Nova redefines AI boundaries, offering cutting-edge solutions that innovate, streamline, and elevate possibilities for enterprises and individuals alike.