Amazon Launches Nova Foundation Models to Set New Standards in AI

Amazon, the well-known American multinational technology company, hosts an AWS event called AWS re: Invent every year. This year, Amazon has launched new family foundational AI models under the Nova branding.

As of now, there are a total of four generative AI models under Amazon Nova, known as Amazon Nova Micro, Lite, Pro, and Premier. Amazon CEO stated that Micro, Lite, and Pro versions will be available instantly to the customers, while Premier will be available in the first quarter of 2025. These models are integrated with Amazon Bedrock, a fully managed service that allows you to use robust AI models to develop smart applications.

Let’s thoroughly understand the Amazon Nova Models integrated inside the Amazon Bedrock.

Amazon Nova Micro: With a content length of 128K tokens, Nova Micro is a text-only model that provides the lowest latency responses at a very low cost. The model is suitable for applications that require fast responses, such as summarization, translation, chat, and easy mathematical reasoning and coding.

Amazon Nova Lite: A cost-effective multimodal AI model that processes texts, images, and videos efficiently. The model possesses the capability to process inputs up to 300K tokens, which means it can quickly process 300 images in one go or a 30-minute video.

Amazon Nova Pro: This highly capable multimodal AI model provides an optimum balance of accuracy, speed, and cost for most tasks. It can also process 300k input tokens in images, texts, or videos.

Amazon Nova Premier: It is the most capable multimodal AI model for complex reasoning tasks. However, Amazon is promoting it as a teacher model for building custom models instead of using the model itself. 

All three models, Lite, Pro, and Premier, can interpret text, audio, and images. They are also multilingual and support up to 200 languages. These models are recommended for tasks like summarizing charts, digesting documents, meetings, and diagrams.

Apart from the text models, Amazon even released two other models, Canvas and Reel. Let’s have a brief discussion about them.

Amazon Nova Canvas: A cutting-edge image generation model that can generate professional-grade images from text or image prompts, outshining the leading model, OpenAI DALL-E3, and stable diffusion. The model comes with an image editing and customization feature. Lastly, the platform has features like watermarking and content moderation.

Amazon Nova Reel: A state-of-the-art video generation model. It generates stunning videos from text and images and is best for advertising and marketing purposes. As of now, the model enables you to generate six-second videos. However, they will also introduce plans that can generate videos up to 2 minutes soon. It gives users the ability to control the style and pacing of the videos. 

How Nova is Revolutionizing Customer Experiences?

Amazon AI models are designed to provide a vast number of benefits to customers. They are as follows:

Enhanced Intelligence: Amazon Nova models deliver best-in-class emotional intelligence across numerous tasks, which makes them ready for multiple applications.

Cost-Effective: The Amazon Nova models are built in a way that makes them a minimum of 75% less expensive compared to the best-performing models in the respective intelligence classes. 

Fast Performance: These multimodal AI models are the fastest in their respective classes, offering low-latency responses, which is essential for real-time applications.

Multimodal Capabilities: Models can easily interpret and provide responses in multiple data types, such as text, images, and videos. Hence, they can be utilized to generate extensive and contextually relevant output.

Retrieval Augmented Generation: Nova models work best with RAG, delivering grounding responses in an organization’s data for maximum accuracy.

Custom Fine-tuning: The model supports custom fine-tuning and distillation. This enables the customers to deliver a particular knowledge from a larger, highly capable “teacher model” to a smaller and more efficient model. The latest foundational model, Nova, understands everything from their own data, and Amazon Bedrock trains a private fine-tuned model to deliver customized responses.

What’s Next Expected from Amazon Nova?

Amazon is all set to introduce two of the most exciting models in 2025: speech-to-speech and multimodal-to-multimodal. Let’s examine them in detail.

Speech-to-Speech Model

This model will have the potential to interpret streaming speech input in the natural language, understand verbal and non-verbal cues, and deliver human-like output with very low latency.

Multimodal-to-Multimodal

The company will introduce an Amazon Nova Model with multimodal-to-multimodal or any-to-any modality capabilities in mid-2025. This will streamline the development of applications anytime a similar model is utilized to execute different tasks, such as translating content from one format to another, editing content, and powering AI agents that can understand and generate all modalities.

Lastly, stay informed about the rapidly changing world of Artificial Intelligence. Get the latest AI advancements, trends, and insights directly in your inbox. Subscribe to our newsletter to join a network of forward-thinkers and innovators shaping the future.