Amazon has taken a new step in the field of artificial intelligence by introducing Nova Premier, the most powerful model in its Nova family. The company announced that the model can process text, image, and video data, although it does not support audio processing. Nova Premier is now available through Amazon's AI platform, Bedrock.
One of the model’s main advantages is its context length of 1 million tokens, which allows it to analyze approximately 750,000 words at once. According to Amazon, Nova Premier performs well in complex tasks that require deep contextual understanding, multi-step planning, and precise execution across multiple tools and data sources. Internal testing shows the model achieved strong results in knowledge retrieval (SimpleQA) and visual understanding (MMMU) benchmarks.
However, the model falls behind in some areas compared to competitors. It underperforms on coding (SWE-Bench Verified), math, and science benchmarks (GPQA Diamond and AIME 2025) when compared to Google’s Gemini 2.5 Pro. Additionally, Nova Premier is not a “reasoning” model, meaning it cannot take extra time or computing resources to deeply consider and fact-check answers like some other models, such as OpenAI’s o4-mini or DeepSeek’s R1.
The pricing structure for Nova Premier is as follows: $2.50 per 1 million input tokens and $12.50 per 1 million output tokens, which is mid-range compared to rival offerings.
Amazon plans to use Nova Premier mainly for training smaller models through a process called distillation — essentially transferring its capabilities into faster, more efficient models tailored for specific use cases. CEO Andy Jassy stated that Amazon is currently developing over 1,000 generative AI applications and that the company’s AI revenue is growing at a triple-digit rate year over year, contributing to a multi-billion-dollar annual revenue run rate. Nova Premier is presented as a key part of this broader AI growth strategy.
