Amazon Web Services (AWS) has officially unveiled its latest suite of AI models, named Nova 2, aimed at empowering enterprise customers with enhanced customization capabilities. During his keynote address at the AWS re:Invent event, CEO Matt Garman introduced these four innovative models designed to improve functionality and user engagement.
The Nova series originally debuted at last year’s conference, featuring four text-generating models alongside a single image-generating model. Garman highlighted the impressive adoption of these models, stating they now serve tens of thousands of clients, spanning from prominent marketing firms to tech innovators like Blue Origin and Robinhood, as well as new startups such as NinjaTech AI.
The latest additions to the Nova family include:
- Nova 2 Lite: This cost-effective reasoning model excels in everyday tasks by processing text, images, and videos.
- Nova 2 Pro: Aimed at more complex requirements, this agent can handle intricate tasks such as coding by utilizing multi-modal inputs.
- Nova 2 Sonic: This model specializes in speech-to-speech interactions designed for conversational AI applications.
- Nova 2 Omni: A comprehensive model capable of processing images, text, video, and speech, producing both text and images.
In conjunction with the model enhancements, AWS launched Nova Forge, a new service allowing enterprises to create their customized versions of the Nova models, referred to as Novellas, at an annual fee of $100,000. This service supports companies in integrating proprietary data with a choice of pre-trained, mid-trained, or post-trained models.
Garman noted that Nova Forge addresses challenges organizations face when personalizing AI models with their own datasets. He provided an analogy comparing model training to language acquisition, illustrating the complexities of adapting knowledge in AI as data is added over time.
Early adopters of Nova Forge include major companies like Reddit, Sony, and Booking.com, all seeking to leverage AWS’s latest capabilities for their unique operational needs.
