Amazon announces Nova, its own artificial intelligence

Amazon's cloud subsidiary Amazon Web Services (AWS) yesterday announced its own generative artificial intelligence models, joining Microsoft, Google, Meta, OpenAI or even Anthropic in the race for AI. Named Amazon Nova, the model is available in 6 versions, 5 of which are available now. For this launch, Amazon is casting a wide net: the different models range from classic text generation to image creation, including video creation. Here are the six models presented by AWS this December 3, 2024:

The six Nova models currently planned:

• Amazon Nova Micro : this model is the simplest of the six. This is classic text-to-text AI, with the lowest response latencies and a context window of 128,000 tokens.
• Amazon Nova Lite : this version offers a more economical multimodal approach, nevertheless capable of taking into account image, video and text in order to generate text. For example, it can be used to analyze up to 30 minutes of video from a simple query.
• Amazon Nova Pro : this multimodal model seeks a balance between precision, speed and cost, and is established according to Amazon as the most efficient model on certain benchmarks.
• Amazon Nova Prime : planned for early 2025, this model should be dedicated to the most sophisticated tasks (used in the professional context in particular).
• Amazon Nova Canvas : As the name suggests, Canvas focuses on generating images from prompts.
• Amazon Nova Reel : this model plans to generate short videos of up to 6 seconds based on a text query. Amazon promises an evolution towards videos of up to 2 minutes in 2025.

Amazon took the opportunity to announce that the three least advanced versions of Nova are “at least 75% less expensive than the best performing models in their category on Amazon Bedrock, in addition to being the fastest“. Note one thing, however: unlike many competitors, Amazon Nova is currently not available to individuals. Only businesses and developers who are AWS customers can benefit from it at the moment.

In 2025, Amazon plans to release two other models in addition to these 6 models already revealed: one of them should be able to transform a textual, video or audio request into a response to the chosen form, textual visual or audio also. Finally, another speech-to-speech model should join the ranks of Amazon's artificial intelligences.