Amazon.com Inc (NASDAQ:AMZN) has unveiled a suite of new AI foundation models under the "Nova" branding.
The initial offerings include three "understanding" models: Micro, optimized for speed and cost; Lite, a low-cost multimodal model for images, video, and text; and Pro, a high-end multimodal model.
Amazon is also developing Nova Premier, a sophisticated multimodal model for complex reasoning, expected in early 2025.
Content-generation models include Nova Canvas for image creation and Nova Reel for video generation, both featuring watermarking for responsible AI use.
Amazon said it plans to launch a speech-to-speech model and a native multimodal-to-multimodal model in late 2025.
Announced at the AWS re:Invent conference in Las Vegas, Amazon also revealed plans for the world’s largest AI compute cluster, built with Trainium 2 chips in collaboration with Anthropic.
This $8 billion investment aims to enhance Anthropic's model development and deployment capabilities.
The company also announced progress on an AI-powered Alexa though its launch has been pushed back to 2025.