Nova 2.0 Lite (high) refers to the high-reasoning configuration of Amazon's second-generation Lite-tier multimodal foundation model. Released in December 2025, it is part of the Nova 2.0 family designed to balance high-speed inference with advanced logical reasoning. This specific configuration utilizes the model's Extended Thinking feature at its highest effort level, enabling the model to perform deep systematic analysis and multi-step problem solving before delivering a final response.
Natively multimodal, Nova 2.0 Lite processes text, images, and video inputs. The model is built on a decoder-only Transformer architecture and features a significant 1,000,000-token context window. This allows for the comprehensive analysis of long-form data, such as documents over 400 pages, extensive software repositories, or videos up to 90 minutes in length, all within a single prompt.
In technical evaluations, the model has demonstrated strong performance in high-risk and complex domains. According to Amazon, Nova 2.0 Lite achieves results comparable to or exceeding other specialized models such as Claude Haiku 4.5 and GPT-5 Mini across standard benchmarks. It also incorporates specific safety optimizations under the Frontier Model Safety Framework (FMSF), assessing and mitigating risks in areas like cybersecurity and specialized scientific knowledge.
Beyond its reasoning capabilities, the model supports integrated tool use, including a built-in code interpreter and web grounding. It is typically utilized for agentic workflows, complex document synthesis, and automated business processes where thorough reasoning is required without the computational overhead of larger frontier models like Nova Premier.