Amazon's Nova AI Models Are Here to Rival ChatGPT and Google ...
A bit late to the party, but Amazon has introduced its first series of foundation AI models at the AWS re:Invent 2024 event. Amazon has developed a family of AI models called Nova that rivals OpenAI’s GPT-4o, Google’s Gemini 1.5 Pro, and its own partner Anthropic’s Claude 3.5 Sonnet (New) model.
About Nova AI Models
There are a total of six models under the Nova family: Nova Micro, Nova Lite, Nova Pro, Nova Premier, Nova Canvas, and Nova Reel. Nova Micro is the smallest text-only model; Nova Lite is a low-cost multimodal model that can process images, videos, and texts; Nova Pro is a capable multimodal model; Nova Premier is the most powerful model designed for complex reasoning tasks (coming in Q1 2025); Nova Canvas can generate AI images and Nova Reel can produce AI videos.
Basically, with Nova AI models, Amazon has covered all the use cases of Generative AI. The Nova models will be available on Amazon Bedrock via its API for developers and business customers to power new AI experiences and applications.
Training and Inferencing
Amazon trained the Nova models on its custom Trainium chips and the inferencing is done on its Inferentia chips. Apparently, Amazon didn’t rely on Nvidia GPUs to train its foundation models, similar to Google’s TPUs.
Integration and Benchmarks
Amazon is going to bring Nova-powered AI experiences through Rufus in the shopping app and Alexa, its voice assistant. As for benchmarks, well, Amazon’s Nova Pro model doesn’t outrank flagship models like Claude 3.5 Sonnet (New) and GPT-4o, but it’s very competitive.
Amazon says its most capable Nova Premier model is coming in the first quarter of 2025. Moreover, customers can finetune models based on their proprietary data.
So what do you think about Amazon’s new Nova AI models? Let us know your thoughts in the comments below.
