Amazon announced a series of new AI foundational models under the "Nova" brand at the AWS re:Invent conference, which will be available on AWS as part of the Amazon Bedrock model library.
Currently, Amazon has launched three "understanding" models:
- Amazon Nova Micro, a text model optimized for speed and cost.
- Amazon Nova Lite, a highly cost-effective multimodal model capable of processing images, videos, and text, and generating text outputs.
- Amazon Nova Pro, a powerful multimodal model.
Amazon is also developing the Amazon Nova Premier model, which is said to be its most powerful multimodal model, specifically designed for handling complex reasoning tasks, and is expected to be released in early 2025.
Additionally, Amazon introduced content generation models: Amazon Nova Canvas for image generation and Amazon Nova Reel for video generation. These models feature watermarking to promote responsible AI use. Amazon showcased an example advertisement for a fictional pasta brand created using Nova Reel.
Amazon plans to launch voice-to-voice models and "native multimodal-to-multimodal" models later in 2025.
At the re:Invent conference, Amazon also announced a collaboration with Anthropic (with an $8 billion investment) to build a massive AI computing cluster based on its Trainium 2 chips. Upon completion, this cluster is expected to be the world's largest AI computing cluster, enabling Anthropic to develop and deploy its future models.
Like other major tech companies, Amazon is racing to launch new AI products and features to maintain its lead over emerging companies like OpenAI. A potential advantage for Amazon is that its AWS already supports vast internet infrastructure, and large enterprises may prefer Amazon's AI services due to the company's established trustworthy reputation in the industry.
Notably, Amazon is developing an AI-driven new Alexa, though the release of the voice assistant originally scheduled for this fall has been postponed to next year.
