Welcome to "The AI Economy," a weekly newsletter by Ken Yeung on how AI is influencing business, work, society, and technology. Subscribe now to stay ahead with expert insights and curated updates—delivered straight to your inbox.
Months after Amazon Chief Executive Andy Jassy first introduced the world to the Nova foundation model family, the company has released what might be its top-of-the-line variant, Amazon Nova Premier. Billed as Amazon’s “most capable model,” it can process text, images, and video (though not audio yet) and tackle complex, multi-step tasks that demand precise coordination across different tools and data sources. Developers interested in Amazon Nova Premier can access it on Amazon Bedrock today.
What separates Nova Premier from its siblings—Nova Pro, Lite, and Micro—is that it serves as a teacher for model distillation, enabling developers to create specialized versions for specific use cases. It has a context length of one million tokens, allowing it to handle lengthy documents (400+ pages), videos (90+ minutes), or codebases. By contrast, that’s significantly larger than Pro (300,000), Lite (300,000), and Micro (128,000).
Amazon states that Nova Premier has been optimized for tasks its customers care about, including code generation, Retrieval-Augmented Generation, video and document understanding, function calling, and agentic interactions.
When compared to Nova Pro, the model most likely used by developers, Premier outperforms across a wide range of benchmarks that span text intelligence, visual intelligence, and agentic workflows.
Subscribe to The AI Economy
Amazon also points out that Premier performs on par with the industry’s top “non-reasoning models,” and matches or outperforms others in its intelligence tier on about half of the benchmarks. The distinction matters: Unlike Nova Pro, Premier lacks reasoning by design. It’s meant to serve as a teaching model rather than a user-facing one. It’s a generalist built for broad, versatile capabilities rather than deep cognitive tasks.
Other distillation models that might be comparable to Nova Premier include those from OpenAI’s GPT-4, Anthropic’s Claude 3 Opus, Google’s Gemini 1.5 Pro, Meta’s LlaMA 3-70B, and Mistral’s Mixtral.
Nova Premier is available on Amazon Bedrock in the U.S. East (Northern Virginia and Ohio) and U.S. West (Oregon) regions. Prices start at $0.0025 per 1,000 input tokens and $0.0125 per 1,000 output tokens.
Featured Image: Amazon introduces the first four Nova variations at the AWS re:Invent conference on Dec. 3, 2024. Photo credit: Ken Yeung
Subscribe to “The AI Economy”
Exploring AI’s impact on business, work, society, and technology.
Leave a Reply
You must be logged in to post a comment.