AWS subscribers now have access to generative AI models comparable to GPT-4o. On December 3, during the AWS re:Invent event in Las Vegas and online, AWS announced six new model sizes for different use cases for the new Amazon Nova family.
“With approximately 1,000 generative AI applications running within Amazon, we wanted to get a bird’s-eye view of what application builders are still working on,” said Rohit Prasad, senior vice president of Amazon Artificial General Intelligence. stated in the press release.
“Our new Amazon Nova model addresses these challenges for both internal and external builders, delivering compelling intelligence and content generation, while improving latency, cost-effectiveness, customization, acquisition augmentation generation (RAG), and agent capabilities. The goal is to make meaningful progress.”
Learn more about innovation
What is Amazon Nova?
Amazon Nova is a series of generative AI foundation models available on the Amazon Bedrock AI hosting service on AWS. Organizations can currently try three size options:
Amazon Nova Micro is a text-only model with a fast response time of 210 output tokens per second. Amazon claims it will perform better than Meta’s Llama 3.1 8B and Google’s Gemini 1.5 Flash-8B. Nova Micro is targeted at applications requiring relatively low cost and fast response. Amazon Nova Lite is another smaller model in the Nova family. Unlike Micro, it can analyze either images, videos, or text input. Comparable to OpenAI’s GPT-4o mini, Nova Lite is aimed at quick summarization and interpretation of charts and video presentations. Amazon Nova Lite’s ability to understand images on a computer screen and make function calls makes it suitable for some semi-autonomous chaining operations used for “AI agent” tasks. Amazon Nova Pro is a mid-range model. Amazon says it is faster, more accurate, and less expensive than OpenAI’s GPT-4o and Google’s Gemini 1.5 Pro. Nova Pro can interpret text or images and supports agent workflows.
Once customers have a Nova model, they can fine-tune the model based on their own data.
In addition to size options, organizations can also choose between image-generated models (Amazon Nova Canvas) and video models (Amazon Nova Reels). Both of these are aimed at creating “studio quality” content.
Nova Canvas creates images based on text or image prompts. Amazon says it includes safety features such as watermarks and content guardrails. Nova Reels creates six-second videos, and Amazon plans to increase video length to two minutes “in the coming months.”
See: AI regulation underway in Australia, with commission calling for large-scale models from OpenAI, Meta and Google to be counted as ‘high risk’.
What’s next?
The fourth model in the Nova series, the Nova Premier, will not go on sale until the first quarter of 2025. Amazon expects Nova Premier to bring multimodal (video, image, or text-to-text) interpretation and a large library of data available to organizations. Use it to train other models.
Amazon also plans to add models that can respond naturally to voice conversations. We also work on multimodal versus multimodal models for interpreting and outputting text, images, video, or audio.
While it’s still too early to see how Nova will compete with rivals like OpenAI, Google, and Meta, Amazon has acquired one of SAP’s key partners to offer models on its AI core platform.
TechRepublic is covering AWS re:Invent remotely.
