Amazon Nova 2 Lite is a multimodal foundation model designed for speed and cost-effective reasoning across high-volume enterprise workloads. Released in December 2025 as part of the Nova 2 family, it supports diverse input modalities including text, images, videos, and documents. The model is positioned as a pragmatic generalist, balancing performance and efficiency for tasks such as document synthesis, customer service automation, and large-scale data extraction.
A defining feature of the model is its extended thinking capability, which provides developers with explicit control over the model's internal reasoning process. Users can toggle this mode and select from different "thinking budgets" (low, medium, or high effort) to prioritize either rapid responses or deep, systematic analysis for complex problems. This allows the model to handle multi-step tasks like mathematical proofs and software architecture design more effectively than standard lightweight models.
Technical Capabilities
Nova 2 Lite features a 1 million-token context window, enabling it to process extensive datasets such as 400-page documents or 90-minute videos in a single prompt. It is equipped with built-in tools for web grounding, which allows it to retrieve and cite public information, and a code interpreter for executing and verifying code within its workflow. These tools, combined with high reliability in function calling, make it suitable for complex agentic interactions.
The model serves as the foundational engine for Amazon Nova Act, a service dedicated to building AI agents for browser-based automation. By utilizing a version of Nova 2 Lite specifically optimized through reinforcement learning for web environments, Nova Act can perform live browser tasks—such as navigating dashboards and filling forms—with approximately 90% operational reliability.