Amazon has launched Nova Premier, its most powerful AI model to date, as part of the growing Nova family.
Available through Amazon Bedrock, the company’s AI model development platform, Nova Premier can process text, images, and videos, although it currently lacks support for audio.
Amazon claims the model excels at complex tasks involving deep contextual understanding, multi-step planning, and seamless execution across various tools and data sources. With a context window of 1 million tokens—equivalent to around 750,000 words—Nova Premier is built for large-scale analysis and task handling.
While Nova Premier lags behind rivals like Google’s Gemini 2.5 Pro in coding benchmarks (SWE-Bench Verified) and scientific reasoning (GPQA Diamond, AIME 2025), it outperforms in knowledge retrieval and visual comprehension, according to Amazon’s internal testing.
The model is priced competitively on Bedrock at $2.50 per 1M input tokens and $12.50 per 1M output tokens, on par with Gemini 2.5 Pro. However, it’s important to note that Nova Premier is not a reasoning model, meaning it doesn’t pause to verify or refine answers like OpenAI’s o4-mini or DeepSeek’s R1.
Amazon is positioning Nova Premier as a key tool for distillation — transferring its capabilities into lighter, task-specific AI models. With CEO Andy Jassy highlighting that Amazon is building over 1,000 generative AI apps and seeing triple-digit AI revenue growth, Nova Premier marks another strategic step in Amazon’s expanding AI ambitions.
