Amazon launches Nova Premier, its most advanced AI model yet
What's the story
Amazon has unveiled its latest artificial intelligence (AI) model, the Nova Premier. The tech giant claims this new addition to the Nova family is its most advanced AI model yet. The Nova Premier is designed to process text, images, and videos (excluding audio), and is accessible on Amazon Bedrock - the company's AI model development platform.
Model features
Nova Premier's unique capabilities and performance
The Nova Premier is built for complex tasks that require a deep understanding of context, multi-step planning, and precise execution across multiple tools and data sources. It features a context length of one million tokens, enabling it to analyze some 750,000 words at once. However, it fell short on some benchmarks against leading AI models from competitors like Google.
Benchmark results
Nova Premier's performance on benchmarks and pricing
On the SWE-Bench Verified coding test, the Nova Premier lagged behind Google's Gemini 2.5 Pro. It also scored poorly on GPQA Diamond and AIME 2025 benchmarks assessing math and science knowledge. Despite these shortcomings, Amazon's internal testing showed the model excelled in knowledge retrieval and visual understanding tests like SimpleQA and MMMU. In terms of pricing, it costs $2.50 per one million tokens fed into the model, and $12.50 per one million tokens generated by it—similar to Gemini 2.5 Pro.
Strategic focus
Nova Premier's role in AI development and Amazon's growth strategy
Amazon has clarified that Nova Premier isn't a "reasoning" model like OpenAI's o4-mini or DeepSeek's R1. This means it cannot evaluate and fact-check its answers to questions. The tech giant is promoting Nova Premier as ideal for "teaching" smaller models via distillation—transferring its capabilities for specific use cases into faster, more efficient packages. CEO Andy Jassy highlighted that Amazon sees AI as increasingly central to its growth strategy, with over 1,000 generative AI applications under development.