Amazon doubles down on AI with next-gen 'Nova' models

Amazon has unveiled a caller acceptable of artificial quality (AI) instauration models called 'Nova', that connection state-of-the-art quality crossed a wide scope of tasks, on with industry-leading terms performance.

These AI models volition beryllium added to the Amazon Bedrock exemplary room connected AWS, the institution informed during the yearly 'AWS re:Invent' here.

“Inside Amazon, we person astir 1,000 generative AI applications successful motion, and we've had a bird's-eye presumption of what exertion builders are inactive grappling with,” said Rohit Prasad, SVP of Amazon Artificial General Intelligence.

Our caller Amazon Nova models are intended to assistance with these challenges for interior and outer builders, and supply compelling quality and contented procreation portion besides delivering meaningful advancement connected latency, cost-effectiveness, customization, Retrieval Augmented Generation (RAG), and agentic capabilities, Prasad explained.

According to the company, Amazon Nova Micro was recovered to beryllium adjacent oregon amended than some Meta LLaMa 3.1 8B connected each 11 applicable benchmarks, and Google Gemini 1.5 Flash-8B connected each 12 applicable benchmarks.

With Amazon Nova Micro's industry-leading velocity of 210 output tokens per second, it is perfect for applications that necessitate accelerated responses, said the company.

In his keynote address, AWS CEO Matt Garman highlighted however the institution is delivering caller innovations successful generative AI, including caller Trainium2 instances, Trainium3 chips, Amazon Nova instauration models and more.

The institution said it volition present an Amazon Nova speech-to-speech exemplary successful the archetypal 4th of 2025.

The exemplary is designed to alteration conversational AI applications by knowing streaming code input successful earthy language, interpreting verbal and non-verbal cues (like code and cadence), and delivering earthy human-like, back-and-forth interactions with debased latency.

Amazon is besides processing a caller exemplary that tin instrumentality text, images, audio, and video arsenic input, and make outputs successful immoderate of these modalities.

This Amazon Nova exemplary with autochthonal multimodal-to-multimodal – oregon “any-to-any” modality capabilities – volition beryllium introduced mid-year 2025.

AWS besides announced the adjacent procreation of Amazon SageMaker, unifying the capabilities customers request for accelerated SQL analytics, petabyte-scale large information processing, information exploration and integration, exemplary improvement and training, and generative AI into 1 integrated platform.