Amazon Announces Nova: A New Family of Multimodal AI Models

Amazon has introduced a new generation of foundation models, dubbed Amazon Nova, designed to revolutionize the way applications handle multiple data modalities. This new family of models, which will be available mid-year 2025, promises to address the challenges faced by application builders within Amazon and external developers.

Key Features of Amazon Nova

Amazon Nova includes four state-of-the-art models tailored to various task requirements:

• Amazon Nova Micro: A text-only model optimized for low latency and very low cost. It supports context lengths of up to 128K input tokens.

• Amazon Nova Lite: A highly cost-effective multimodal model that is lightning fast for processing image, video, and text inputs. It supports context lengths of up to 300K input tokens.

• Amazon Nova Pro: A highly capable multimodal model, offering the best combination of accuracy, speed, and cost for a wide range of tasks. It also supports context lengths of up to 300K input tokens.

• Amazon Nova Premier: The most advanced multimodal model, designed for complex reasoning tasks and ideal for use as a teacher to distill custom models. This model will be available in Q1 2025.

Each of these models supports over 200 languages, making them versatile across different linguistic and cultural contexts.

Integration with Amazon Bedrock

All Amazon Nova models are seamlessly integrated with Amazon Bedrock, a fully managed service that provides access to high-performing foundation models (FMs) from leading AI companies and Amazon through a single API. This integration allows customers to easily experiment with and evaluate Amazon Nova models alongside other FMs to determine the best model for their application.

Innovative Applications

Amazon also emphasizes the integration of its new models with various creative tools. For instance:

• Amazon Nova Canvas: A state-of-the-art image generation model designed to produce high-quality images from text inputs.

• Amazon Nova Reel: A state-of-the-art video generation model allowing users to create compelling videos from text and images.

Future Developments

Amazon plans to introduce additional models in the future. One upcoming development includes a speech-to-speech model set to be released in the first quarter of 2025. This model aims to transform conversational AI by understanding streaming speech input, interpreting verbal and non-verbal cues, and delivering human-like back-and-forth interactions with low latency. Another model under development will feature native multimodal capabilities, allowing for text, images, audio, and video inputs to be processed and generate outputs in any of these modalities.

Benefits for Customers

The new Amazon Nova models offer exceptional intelligence and speed while being cost-effective. They have been designed to be highly customizable, supporting fine-tuning with customers’ proprietary data. This customization ensures tailored responses that leverage the specific knowledge from an organization’s data.

Impact on Agentic Applications

Amazon Nova models are optimized for agentic applications, which require interacting with an organization’s systems through multiple APIs to execute complex tasks. The models also support Retrieval Augmented Generation (RAG), enabling customers to ground responses in an organization’s own data.

Amazon’s introduction of Amazon Nova represents a significant step forward in the field of AI, addressing the pressing needs of application builders across various industries. With its robust integration capabilities, innovative applications, and commitment to customization, Amazon Nova is poised to transform the way organizations leverage artificial intelligence.