Everything you need to know about Amazon Nova

Amazon Nova offers models tailored to different needs. Nova Micro is a text-only model that delivers the lowest latency responses at very low cost. Nova 2 Lite is a fast, cost-effective reasoning model for everyday workloads. It is ideal for customer service chatbots, document processing, and business automation, with standout capabilities in processing documents, extracting information from videos, generating code, providing accurate answers, and automating multi-step agentic workflows. Nova 2 Pro is Amazon’s most intelligent reasoning model, designed for highly complex tasks like agentic coding, long-range planning, and sophisticated problem-solving where the highest accuracy is essential.

Nova Forge empowers organizations to customize and train optimized variants of Nova models by blending their proprietary data with Nova's frontier capabilities. The service provides exclusive access to pre-trained, mid-trained, and post-trained Nova model checkpoints so customers can mix their proprietary data with Amazon Nova-curated datasets at every stage of training. The result is a customized model that combines Nova's full knowledge and reasoning power with a deep understanding of each organization's specific business.

Beyond model checkpoints, Nova Forge offers reinforcement learning "gyms"—synthetic environments where models learn from simulated scenarios. It also offers synthetic data-based distillation for creating smaller, faster models that maintain their intelligence at lower cost, and a responsible AI toolkit for safety controls.

Organizations from cutting-edge medical research startups to leading global media brands are using Nova Forge to build custom models and drive efficiencies and innovation in their businesses. Once customers create and train custom models tailored to their needs with Nova Forge, they can deploy them on Amazon Bedrock with enterprise-grade security, scalability, and data privacy.

Available as a service on AWS, Nova Act enables the building and deploying highly reliable AI agents that automate browser-based tasks, delivering 90% reliability on early customer workflows. Powered by a custom Nova 2 Lite model trained through reinforcement learning, Nova Act excels at UI-based workflows like updating data in customer relationship management (CRM) systems, testing website functionality, or submitting health insurance claims. Developers can prototype agents in minutes using a no-code playground, refine them in familiar IDEs like VS Code, and deploy them to AWS with comprehensive management tools. Organizations like Hertz, 1Password, Sola Systems, and Amazon Leo are using Nova Act to accelerate software delivery by 5x and automate hundreds of thousands of workflows per month.

Nova 2 Sonic is Amazon's updated speech-to-speech model that unifies text and speech understanding and generation for real-time, human-like conversational AI. It features expanded multilingual support with expressive voices, higher accuracy, and a one-million token context window for sustained interactions, with seamless switching between voice and text. The model handles tasks asynchronously, letting users continue natural conversations—even switching topics—while actions complete in the background.

Amazon Nova Canvas and Nova Reel enable customers to create high-quality visual content at scale. Nova Canvas generates studio-quality images from text prompts, supporting use cases from marketing campaigns to product visualization. Nova Reel produces professional-grade video content, allowing organizations to create everything from social media clips to training materials without extensive production resources. For multimodal workflows, Nova 2 Omni processes text, images, videos, and speech inputs while generating both text and images—enabling teams to analyze content across multiple formats and generate campaign elements in one workflow.

Amazon Nova Multimodal Embeddings is a state-of-the-art multimodal embedding model for agentic retrieval-augmented generation (RAG) and semantic search applications, available on Amazon Bedrock. It is the first unified embedding model that supports text, documents, images, video, and audio through a single model, enabling crossmodal retrieval with leading accuracy. Embedding models convert text, visual, and audio inputs into numerical representations that AI systems can compare, search, and analyze. This unlocks insights from unstructured data across multiple content types, powering use cases such as crossmodal search across mixed-modality content, searching with reference images, and retrieving visual documents—eliminating the need to build complex crossmodal embedding solutions or restrict use cases to a single content type.

All Amazon Nova models are available to AWS customers in Amazon Bedrock, a fully managed service that makes high-performing foundation models available through a single API. In addition, nova.amazon.com provides an easy way to experiment with Nova models—whether you’re testing model behavior, exploring new AI capabilities, seeking creative inspiration, or experimenting with agents. Start building for free at nova.amazon.com/dev.