Key Points
- AWS introduced four new Nova 2 models: Lite, Pro, Sonic and Omni.
- Nova 2 Lite focuses on cost‑effective reasoning tasks.
- Nova 2 Pro handles complex, multimodal workloads such as coding.
- Nova 2 Sonic adds speech‑to‑speech capabilities for conversational AI.
- Nova 2 Omni is a multimodal model that processes text, images, video and speech.
- The new Nova Forge service lets enterprises build custom “Novellas” for $100,000 per year.
- Nova Forge offers pre‑trained, mid‑trained and post‑trained model options.
- Early Nova Forge customers include Reddit, Sony and Booking.com.
- AWS aims to help enterprises integrate proprietary data without losing core model reasoning.
New Nova 2 Model Family Expands Multimodal AI Capabilities
During the AWS re:Invent keynote, Amazon Web Services introduced four new models that extend the Nova family. Nova 2 Lite is positioned as a cost‑effective reasoning model, while Nova 2 Pro targets highly complex tasks such as coding, processing text, images, video and speech. Nova 2 Sonic adds a speech‑to‑speech capability for conversational AI, and Nova 2 Omni is a multimodal reasoning and generation model that can handle image, text, video and speech inputs and produce both text and images. The expanded suite builds on the original Nova models announced the previous year and is already being used by a broad range of customers, from large enterprises to innovative startups.
Nova Forge Enables Enterprise‑Level Model Customization
Alongside the model upgrades, AWS announced Nova Forge, a new service that allows cloud customers to create their own customized versions of Nova models—referred to as “Novellas”—for an annual fee of $100,000. The service offers access to pre‑trained, mid‑trained, or post‑trained models, giving companies the ability to fine‑tune the AI on proprietary data while preserving core reasoning capabilities. Early adopters of Nova Forge include Reddit, Sony and Booking.com.
Strategic Implications for AWS and Its Customers
The launch reflects AWS’s strategy to deepen its AI portfolio and address enterprise concerns about integrating proprietary data with large‑scale models. By providing a pathway for model customization, AWS aims to reduce the “forgetting” effect that can occur when models are retrained with new data, a challenge likened to learning a new language later in life. The expanded Nova lineup and Nova Forge service position AWS to compete more aggressively in the generative AI market, offering flexible, multimodal solutions that cater to a wide spectrum of business needs.
Source: techcrunch.com