Nova 2 Lite

Nova 2 Lite is designed as a fast and cost-effective reasoning model that brings together a large context window with multimodal understanding. It accepts text, images, video, and documents as input and produces text output, and supports configurable extended thinking, web grounding, and code execution features aimed at agentic workflows. A 1M token context window allows it to handle long documents, multi-turn reasoning, and retrieval-augmented generation tasks without losing track of prior information. The model is positioned within the broader Nova 2 family to balance speed, cost, and intelligence, making it well suited for high-volume production use cases such as RAG pipelines, autonomous coding agents, and conversational assistants that need to reason across varied media. The second-generation design focuses on practical strengths that show up in real deployments, including low-latency responses for interactive agents and the ability to generate very long outputs. An AWS re:Invent demonstration highlighted its use in autonomous code-improvement agents built with the Strands framework and GitHub's Model Context Protocol, where it analyzed issues and orchestrated multi-step actions. Its multimodal reasoning, tool use, and temperature control make it a flexible generalist for tasks ranging from multimodal understanding to agentic automation, while its price-performance tier targets everyday production workloads rather than premium frontier reasoning.

As a member of the Nova 2 family, Nova 2 Lite builds on Amazon's continued push into agentic AI, complementing newer offerings like Nova Forge, which lets organizations create custom Nova variants by infusing proprietary data during an open training process, and Nova Act, which targets browser-based UI automation. Nova 2 Lite is positioned as the accessible, high-throughput workhorse of that lineup, optimized for cost efficiency while still delivering strong reasoning and tool-using capabilities. Its combination of a 1M token context, extended thinking controls, and native support for text, image, and video inputs makes it a natural fit for forward-looking applications such as long-context document analysis, code-generating developer tools, and multimodal assistants that need to reason over diverse content while keeping per-token costs low for production scale.