Amazon unveiled a series of large AI language models under the Nova brand at its AWS re:Invent conference that will be available to AWS cloud platform customers in the Amazon Bedrock model library.
Image source: BoliviaInteligente / unsplash.com
There are currently three “understanding” AI models available to customers:
- Amazon Nova Micro is a “optimized for speed and price” model for working with text;
- Amazon Nova Lite is a “very low-cost” multimodal model that generates text by accepting not only text, but also images and video as requests;
- Amazon Nova Pro is the “most capable” multimodal model.
The Amazon Nova Premier model is in the process of training – it, according to the developer, will gain the ability to reason and solve the most complex problems. The company also introduced two “creative” systems: Amazon Nova Canvas generates images, and Amazon Nova Reel generates videos. The materials they create receive appropriate tags in the metadata – a measure aimed at “responsible use of AI.” In 2025, Amazon will release two more models: one designed for voice interaction and one focused on multimodal requests and responses.
At the AWS re:Invent event, Amazon announced that it is partnering with Anthropic to build a huge computing cluster for AI systems on Trainium 2 accelerators, where the startup, in which it has invested $8 billion, will be able to create and deploy its projects. Earlier, Apple took part in the conference, which, as it turned out, also uses chips developed by Amazon in working with AI. Amazon continues to work on an updated version of the Alexa voice assistant – according to unofficial data, its release was planned for last fall, but now it has probably been postponed to next year.