Amazon Unveils Nova, a New Family of Multimodal AI Models
AWS introduces Nova, a new suite of generative AI models that process text, images, and video across four tiers of capability. The models promise fast performance, low costs, and massive context windows for cloud customers.
Amazon Web Services unveils Nova, a new family of multimodal generative AI models available across four text-generating tiers: Micro, Lite, Pro, and Premier. While the first three tiers launch immediately for AWS customers, the highly capable Premier model arrives in early 2025. In addition to the text generators, the Nova lineup includes Nova Canvas for image generation and Nova Reel for video creation.
The text-generating models cater to different needs by offering varying levels of speed, accuracy, and cost. Micro provides the lowest latency for text-only tasks, while Lite, Pro, and Premier handle text, image, and video inputs for complex workflows like document digestion and meeting summarization. AWS designs the Premier tier specifically to act as a teacher model for creating fine-tuned, custom AI solutions rather than serving as a standalone tool.
These models feature substantial context windows, with Lite and Pro supporting 300,000 tokens to process hundreds of thousands of words or 30 minutes of video. Amazon CEO Andy Jassy highlights that the Nova lineup ranks among the fastest and most cost-effective models in their class. Developers access and fine-tune the models through AWS Bedrock, where the systems are optimized for complex, multi-step agent behaviors.