Amazon Unveils Nova, a New Generation of AI Models
Amazon has introduced Nova, a new generation of artificial intelligence (AI) models, consisting of five models that can process and understand text, images, and videos, as well as generate creative content. These models are now available on Amazon Bedrock, a platform for developers of generative AI applications.
Key Features of Nova Foundational Models
Nova foundational models (FM) have the capacity to process text, images, and videos, as well as understand graphics and other multimedia content. The five models that make up Nova are:
- Amazon Nova Micro: a text-only model that offers lower latency responses at a low cost
- Amazon Nova Lite: a model that has greater agility to process image, video, and text inputs
- Amazon Nova Pro: a highly capable multimodal model that combines precision and speed for a wide range of tasks
- Nova Canvas: a state-of-the-art image generation model
- Nova Reel: a state-of-the-art video generation model
Capabilities and Applications
All Nova models support a wide range of tasks in 200 languages and multiple modalities. They can also be adjusted based on a range of preferences, allowing them to “learn what customers care about most from their own data (including text, images, and videos), while Amazon Bedrock trains a custom model that will provide personalized responses.”
The new models are designed to help internal and external developers face challenges in application development, providing “compelling intelligence and content generation” and offering significant progress in latency, profitability, and personalization.
Pricing and Future Developments
Amazon Nova Micro, Amazon Nova Lite, and Amazon Nova Pro are at least 75 percent cheaper than the top-performing models in their respective intelligence classes on Amazon Bedrock.
Amazon has also announced that it will present two new Amazon Nova models in 2025: a speech-to-speech conversion model and a native multimodal model, which will be able to process text, images, audio, and video, both as input and output values.