Amazon introduced a new generative AI model, Nova Premier, which the company calls the most powerful in its Nova lineup. The model is available on the Amazon Bedrock service and can work with text, images, and video, but does not support audio. Nova Premier stands out for its ability to perform complex tasks that require deep contextual understanding, multi-step planning, and precise execution of actions involving various tools and data sources.
The model has a context window of up to one million tokens, which is roughly equivalent to seven hundred and fifty thousand words. This enables the analysis of large volumes of information in a single request. According to Amazon, Nova Premier demonstrates strong results in knowledge retrieval and visual understanding tests, such as SimpleQA and MMMU.
On the Bedrock service, the cost of using Nova Premier is two and a half US dollars per million input tokens and twelve and a half dollars per million output tokens, which is comparable to the prices of competing models. At the same time, Nova Premier does not belong to the models with advanced reasoning — it cannot additionally verify its own answers or spend more time processing complex questions.
Amazon positions Nova Premier as an optimal solution for training smaller models via distillation, that is, transferring its capabilities into simplified and faster variants for specific tasks. The company is actively developing its generative AI direction and sees it as an important element of its growth strategy, reporting triple-digit annual revenue growth in this segment.
