Nova Pro

Nova Pro is the accuracy-focused tier in Amazon's first Nova generation, built to handle high-stakes multimodal tasks where cheaper models tend to produce unreliable results. Its design centers on three demanding use cases: parsing dense financial reports and regulatory filings with numerical precision, reasoning across large codebases—up to 15,000 lines in a single request—to catch logic errors and assess architectural patterns, and processing combined visual and textual evidence for applications like insurance claim review or manufacturing quality inspection. The model accepts text, images, and video in the same request, maintaining semantic distinctions across complex, structured inputs where a single misinterpreted figure can cascade into downstream errors.

Developed by Amazon's Artificial General Intelligence team as part of a broader family including Lite, Micro, Canvas, and Reel variants, Nova Pro occupies the middle ground between budget-friendly options and frontier capability. Its agentic features include reliable function calling and long-range planning, earning competitive scores on Berkeley's tool-use benchmarks and excelling in retrieval-augmented generation tasks where speed and cost-effectiveness matter alongside accuracy. Head-to-head comparisons show Nova Pro answering RAG questions faster and more cost-effectively than competing models while maintaining matching accuracy, making it a practical choice for orchestration-heavy workflows where agents make dozens of calls per task.