Amazon reveals 6 new AI models

At its re:Invent 2024 conference in Las Vegas, Amazon unveiled a new family of six AI models, which it’s calling the ‘Nova’ series.  

In the family of six, there are four text-generation models: Nova Micro, Nova Lite, Nova Pro, and Nova Premier (which isn’t available just yet but should be in early 2025). And there’s also an image generator, Canvas, and a video generator, Reel.

Nova Micro, as the name suggests, is the smallest model and can only process and respond in text (it can process up to 100,000 words at a time). But it’s been specifically designed for high speed at a low cost and is reportedly the fastest model available, in its size.

Nova Lite is a slightly bigger multimodal model that can process text, plus images and videos quickly, at a “very low cost,” whereas Nova Pro—also multimodal—is “highly capable” and offers a balance of accuracy, speed, and cost over a range of tasks. Both these models can process up to 250,000 words, 15,000 lines of code, and 30 minutes of video footage and are ideal for tasks like digesting documents and summarizing charts, meetings, and diagrams.

Nova Premier is Amazon’s “most capable multimodal model for complex reasoning tasks” but Amazon is positioning this one as more of a ‘’teacher model’’ to help create fine-tuned, custom AI models, rather than to be used on its own.

Canvas can do the usual image creation/editing tasks, based on prompts, like removing backgrounds and changing color schemes and layouts, and Reel can generate videos (via prompts) up to six seconds long (a 2-minute version is on the horizon), which takes around 3 minutes. It can also allow users to adjust the camera motion and add pans, 360-degree rotations, and zoom-in/out effects.  

Although Amazon is chasing the likes of OpenAI, Anthropic, and Google with these new models, they do have the infrastructure in place, powered by AWS, which is trusted by many large companies and may, therefore, be appealing to the public when considering which AI model to use.