In conjunction with its announcement of Nova Forge, a platform for building customized variants of its Nova foundation models, Amazon Web Services Inc. today introduced four new artificial intelligence models under the Nova banner that expand AWS’s generative AI offerings in multimodal reasoning, speech processing and user interface automation.

The additions to the Nova family, announced at the cloud giant’s re:Invent conference today in Las Vegas, are each designed for different levels of reasoning complexity and multimodal processing.

Basic to advanced

Nova 2 Lite is described as a cost-efficient reasoning model intended for everyday workloads. It can process text, images and video, generating text outputs for tasks such as customer service chatbots, document analysis and business automation. The model allows users to control how much step-by-step reasoning it performs to achieve the needed balance of latency and accuracy. Lite includes built-in web grounding and code execution capabilities, enabling it to incorporate current information into its responses.

AWS called the new Nova 2 Pro its most capable reasoning model. It supports text, images, video and speech inputs and is aimed at advanced tasks involving long-range planning, complex instructions or agentic coding. Like Lite, Pro includes web search and code execution features. It can also act as a “teacher” model for distillation, helping customers create smaller variants tailored to specific workloads.

Nova 2 Sonic is a speech-to-speech model that unifies text and voice understanding and generation. It supports real-time conversational interactions in multiple languages while tasks run asynchronously in the background. Its 1-million-token context window is equivalent to about 75,000 lines of code or 1,500 pages of text. Sonic is built for interactive voice systems and integrates with Amazon Connect cloud contact center service, telephony partners and conversational AI frameworks.

Nova 2 Omni is the first Nova model built for full multimodal generation. It supports text, image, video and speech inputs and can generate both text and images. The model is designed to handle large volumes of mixed-media input, such as lengthy documents, videos and audio files, in a single workflow. AWS said Omni eliminates the need to combine multiple specialized models. It can, for example, ingest entire product catalogs and produce multifaceted marketing campaigns from the contents.

The new Nova 2 models are available now. Developers can prototype applications using Nova tools at nova.amazon.com/dev, and enterprises can deploy models on Amazon Bedrock with standard security, privacy and scalability controls.

Photo: Robert Hof/SiliconANGLE