Last year, Amazon announced the new Nova series of state-of-the-art foundation models exclusively for Amazon Bedrock customers. Amazon Nova Micro, Amazon Nova Lite, and Amazon Nova Pro are LLMs that accept text, image, and video inputs and generate text output.
Today, Amazon announced the availability of Nova Premier, its most capable model suitable for complex tasks and as a teacher for model distillation. Amazon claims that Nova Premier can be used in Amazon Bedrock Model Distillation to create cost-effective, low-latency versions of Amazon Nova Pro, Lite, and Micro for certain verticals and specific scenarios.
Nova Premier supports a 1-million-token context window and over 200 languages. In terms of performance, Nova Premier is not a state-of-the-art (SOTA) model. Amazon itself has stated in an official blog post that Nova Premier is comparable to equivalent non-reasoning models on the market and is equal to or better than approximately half of the evaluated benchmarks.
According to Amazon, Nova Premier performs better on complex tasks that require deep contextual understanding, multistep planning, and precise execution across multiple tools and data sources.
A benchmark comparison table in Amazon's whitepaper omits all the latest models from OpenAI and Google, seemingly to make Nova Premier's performance appear more favorable. Additionally, Amazon highlights that Nova Premier is the fastest and most cost-effective model in Amazon Bedrock within its intelligence tier. Outside of AWS, developers may find that OpenAI's GPT-4.1 series and Google's Gemini Flash series offer better intelligence at even more competitive price points.
Nova Premier is now available in Amazon Bedrock in the US East (N. Virginia), US East (Ohio), and US West (Oregon) AWS Regions via cross-Region inference.
0 Comments - Add comment