Introducing Nova Premier! Amazon's Newest Smart AI Model Designed for Complex Tasks

Kattiya Jantas
May 2
3 min read

Amazon has announced the launch of “Nova Premier,” the latest intelligent AI model in the Amazon Nova family. It comes with robust capabilities in understanding text, images, and video (excluding audio), and is now available on Amazon Bedrock—Amazon’s platform for building and deploying AI models.

Nova Premier excels at handling complex, multi-step tasks that require precise planning and tool integration. It supports a context length of up to 1 million tokens, equivalent to about 750,000 words, making it ideal for analyzing large documents or extensive codebases.

While Nova Premier may still lag behind some leading AI models in specific benchmarks such as mathematics, science, or coding (e.g., SWE-Bench Verified), it performs exceptionally well in knowledge retrieval (SimpleQA) and multimodal understanding of images and videos (MMMU)—areas Amazon has chosen to highlight as its key strengths.

Introducing Nova Premier! Amazon's Newest Smart AI Model Designed for Complex Tasks

Key Features of Nova Premier

Supports text, image, and video input
Processes up to 1 million tokens of context
Optimized for complex analysis and multi-step workflows
The fastest and most cost-effective model in its class on Bedrock

Nova Premier Trains Smaller Models to Be Smarter

Beyond its own versatility, Nova Premier can also transfer its capabilities to smaller models such as Nova Micro, Lite, and Pro using a method similar to “teaching by example.” It uses question-answer pairs generated by Premier to train the smaller models.

For instance, Amazon used Nova Premier to train Nova Pro, improving its accuracy in selecting tools and calling APIs by 20%, while maintaining the same speed and low cost.

Nova Premier in Investment Research

Amazon highlights how Nova Premier can be used in multi-agent architectures for investment research. Here, Nova Premier acts as the lead agent overseeing the overall task, while other models like Nova Pro serve as specialized agents gathering data from various sources.

Workflow example:

Nova Premier analyzes a question, such as “What are the investment trends in clean energy?”
It breaks down the question into subtopics: economic indicators, news, or market sentiment.
Assigns Nova Pro agents to collect data for each subtopic.
Aggregates the analysis into a summary report for financial experts.
This approach reduces analysis time and significantly improves accuracy.

Want to Save Even More? Train Nova Micro with Nova Premier

Amazon also recommends using Nova Premier to train Nova Micro for specific tasks—like the investment research use case—without needing large amounts of labeled data. Premier’s outputs can directly serve as training examples.

Steps:

Use Nova Premier to generate synthetic datasets from real user questions.
Train Nova Micro to respond similarly to Premier.
Evaluate for speed and accuracy before deploying.

Nova Premier Pricing

Input processing: $2.50 per 1 million tokens
Output generation: $12.50 per 1 million tokens

This pricing is comparable to competitors like Gemini 2.5 Pro, but Nova Premier offers faster performance and the ability to train other models, giving it a unique edge.

What Early Customers Say About Nova Premier

Slack: “Excellent for interactive analysis—faster and nearly half the cost of competitors.”
Robinhood: “Enables effective multi-agent workflows using small models like Nova Micro. Great value and easier access.”
Snorkel AI: “We're excited to accelerate domain-specific model development using Nova Premier’s training capabilities.”

Nova Premier: The New Heart of AI at Amazon

Nova Premier is more than just another smart model. It’s the central pillar of Amazon’s AI strategy—enabling the development of smaller, smarter, and more cost-efficient models ready for real-world enterprise use. If you're interested, you can request access via the Amazon Bedrock Console.