Amazon Launches Nova Premier: Its Most Advanced AI Model Yet

Amazon Nova Premier

Amazon has just made its most powerful AI model, Nova Premier, generally available through Amazon Bedrock. This new model joins the existing Nova family Nova Lite, Pro, and Micro and is designed for solving complex problems with better speed, accuracy, and cost-efficiency.

What’s Special About Nova Premier?

  • Multimodal Support: It can understand and process text, images, and videos (but not audio).
  • Huge Memory: With a 1 million token context length, it can handle long documents and big code bases.
  • Smart and Fast: Nova Premier is the smartest Nova model yet, and also the fastest and most cost-effective in its class on Amazon Bedrock.

Nova Pro

Used as a Teacher Model

Nova Premier can also be used to "teach" smaller models (like Nova Pro, Lite, or Micro) using model distillation. This means you can create smaller, cheaper, and faster models that still perform like the big one.

For example, when Nova Premier was used to distill Nova Pro for tasks like selecting tools or making API calls, the result was a 20% accuracy boost—while keeping Nova Pro’s lower cost and faster response.

NOva premier

Real-World Example: Financial Research

Nova Premier can act as a “supervisor agent” in a multi-agent AI system. In a finance app, it can break down a complex question like “What are the trends in renewable energy investments?” into smaller steps, assign tasks to smaller agents (like Nova Pro), gather the results, and produce a final report all automatically.

User

How to Use It?

  • Request access to the Amazon Bedrock console
  • Use it with the Bedrock Converse API
  • Available in US East (N. Virginia and Ohio) and US West (Oregon) regions

What Customers Are Saying?

Big names like Slack, Robinhood, and Snorkel AI are impressed with Nova Premier’s capabilities, praising its power, speed, and flexibility for real-world AI solutions.

In Short. Nova Premier is Amazon’s smartest, fastest, and most efficient AI model, perfect for complex tasks, training smaller models, and powering intelligent workflows in real-world applications.