Amazon Web Services is introducing a range of new internally developed AI models, along with a service designed for businesses to create their own personalized versions.
During AWS CEO Matt Garman’s AWS Re:Invent keynote on Tuesday, the cloud giant unveiled Nova 2, a collection of four new AI models joining its existing Nova model series.
The initial iteration of AWS Nova was revealed last year at the company’s yearly technology conference, which included four models for text generation and one for image generation. This year, AWS is enhancing these models and launching an associated service.
“The momentum has been truly remarkable,” Garman stated during his Tuesday keynote. “Nova has grown to be utilized by tens of thousands of customers today, spanning from major marketing firms to technology leaders such as Infosys, Blue Origin, or Robinhood, as well as innovative startups like NinjaTech AI, and today, we are making Nova even better.”
Among the four new models are Nova 2 Lite, a more economical reasoning model. Reasoning AI models are designed to “think” before providing responses, capable of processing text, images, and videos to generate text suitable for routine tasks. Nova 2 Pro is a reasoning agent that can handle text, images, videos, and speech, engineered for “highly complex tasks” like software development.
Nova 2 Sonic is a novel speech-to-speech model intended for use in conversational AI applications. Nova 2 Omni stands as a multimodal reasoning and generation model, adept at processing inputs across images, text, video, and speech, and generating both text and images.
In conjunction with these model enhancements, AWS also unveiled Nova Forge, a new service empowering AWS cloud clients to develop their own cutting-edge versions of AWS Nova models, termed Novellas, for $100,000 annually, as reported by CNBC. This service grants enterprises access to pre-trained, mid-trained, or post-trained models for further customization with their proprietary data.
Garman indicated that this solution aims to resolve challenges encountered when enterprises attempt to integrate their unique data into AI models that have already undergone initial training.
“The more you customize models, the more you add a lot of data in post-training, these models tend to lose some of that interesting core reasoning they acquired earlier,” Garman explained. “It’s somewhat similar to humans learning a new language. When you start very young, it’s relatively easy to pick up, but attempting to learn a new language later in life is significantly harder. Model training exhibits a similar pattern.”
Early adopters of Nova Forge include prominent companies such as Reddit, Sony, and Booking.com.
Stay updated with all of TechCrunch’s reporting from the annual enterprise technology event here.
Sponsored: Watch re:Invent industry streams
Explore the latest announcements on topics ranging from agentic AI and cloud infrastructure to security and more from Amazon Web Services’ flagship event in Las Vegas. This video is brought to you in partnership with AWS.