Amazon has launched Nova Premier, its most powerful model to date, aimed at handling demanding, complex workloads.
It’s now available through Amazon Bedrock, the company’s platform for building and deploying large models.
Unlike earlier versions, Nova Premier is built to process text, images, and videos all at once, with an interesting context window of one million tokens. That means it can interpret and respond to roughly 750,000 words in a single run. Still, it doesn’t support audio.
Amazon says this model is best used for scenarios that involve multiple steps and tools. It’s particularly strong at understanding visual content and pulling up accurate information from vast data sets.
It’s also being positioned as a “teacher” — that is, it can be used to create smaller, faster versions of itself like Nova Pro, Micro, and Lite. These distilled models are cheaper to run and better suited for production environments.
Nonetheless, Nova Premier doesn’t reason like other top-tier models. For example, it won’t stop to cross-check or refine its answers the way OpenAI’s o4-mini or DeepSeek’s R1 do. And on some coding and academic benchmarks, it falls behind Google’s Gemini 2.5 Pro. But it does outperform others on specific tests like SimpleQA and MMMU, which focus on retrieval and image comprehension.
In practical terms, Amazon is selling access at $2.50 for every million input tokens and $12.50 for every million output tokens — putting it in the same pricing bracket as Google’s Gemini 2.5 Pro.
Amazon wants businesses to think long-term. Nova Premier isn’t just a flashy tool; it’s a strategy piece. It’s the backbone for building smarter, leaner models with lower costs and faster speeds. Andy Jassy, Amazon’s CEO, has made it clear: “We’re building more than 1,000 generative AI applications… AI revenue is growing at triple-digit percentages and represents a multi-billion-dollar annual revenue run rate.”
Some early users are already seeing the impact.
“Amazon Nova Premier has been outstanding in its ability to execute interactive analysis workflows, while still being faster and nearly half the cost compared to other leading models in our tests,” said Curtis Allen, senior staff engineer at Slack.
Robinhood’s Head of AI and Data, Dev Tagare, added, “Implementing new solutions built on top of Amazon Nova has helped us with our mission of democratizing finance for all… The intelligence of Nova Premier and what it can transfer to the other models like Nova Micro, Nova Lite, and Nova Pro unlocks multi-agent collaboration at a performance, price, and speed that will make it accessible to everyday customers.”
And from Snorkel AI, Henry Ehrenberg noted, “Accelerating real-world AI deployments—not just prototypes—requires the ability to build models that are specialized for the unique needs of real world applications. We’re excited to see AWS pushing efficient model customization forward with Amazon Bedrock Model Distillation and Amazon Nova Premier.”
Users can access Nova Premier now in three U.S. regions (N. Virginia, Ohio, and Oregon) via cross-region inference. Amazon also offers a site — nova.amazon.com — for customers to test and explore Nova models directly.
