Amazon

“Amazon Nova 2 Lite serves a wide range of potential application domains and offers the following core capabilities: Long Context: Amazon Nova 2 Lite supports a 1M token context window Reasoning across a wide variety of inputs including text, documents, charts, images, and videos: Amazon Nova 2 Lite offers advanced understanding capabilities, enabling deeper insights from multimedia content Reasoning Controls: Amazon Nova 2 Lite is a hybrid reasoning model, so you can turn reasoning on/off depending on the task Reasoning Budget: Amazon Nova 2 Lite supports a reasoning budget which allows you to configure the amount of reasoning needed for tasks by changing the setting to "low", "medium", or "high" Multimodal Reasoning: Amazon Nova 2 Lite will integrate multimodal inputs into its reasoning process, improving performance on tasks requiring understanding of charts, diagrams, or code screenshots Improved Instruction Following: Amazon Nova 2 Lite improves in its inherent instruction following capabilities in comparison to previous generations Sequential Task Decomposition and Function Calling: Amazon Nova 2 Lite will automatically decompose complex problems into manageable sub-tasks System Tools: Amazon Nova 2 Lite provides two different system tools that the model can use when responding to a user's query Amazon Nova Grounding allows the model to access the web and search for live information The Code Interpreter allows the model to perform calculations using a python interpreter Agentic workflows for enabling applications: Amazon Nova 2 Lite facilitates applications that require RAG, API execution, and User Interface (UI) actuation, such as predicting API actions to automate client applications Customizability: Amazon Nova 2 Lite supports Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) which allows you to adapt the model to your business-specific needs, see the Amazon Nova User Guide.”

“In one of the proprietary datasets containing 6.4k prompts designed to elicit harmful responses, such as those related to self-harm and animal abuse, Amazon Nova 2 Lite correctly provides safe responses to over 98% of these prompts.”

“Customers should carefully consider whether or not using RAG will improve the effectiveness of their solution; use of RAG can still result in errors.”

“Our runtime service architecture works as follows: 1/ the model receives a user prompt via the API or Console; 2/ the model filters the prompt to comply with safety, security, and other design goals; 3/ the model may augment the filtered prompt to support user-requested features, for example, grounding to access the web and search for live information; 4/ the model generates a completion; 5/ the model filters the completion for safety and other concerns; 6/ the model returns the final completion.”

“The UK and Republic of Korea governments announced that the following organisations have agreed to the Frontier AI Safety Commitments: Amazon Anthropic Cohere Google G42 IBM Inflection AI Meta Microsoft Mistral AI Naver OpenAI Samsung Electronics Technology Innovation Institute xAI Zhipu.ai The following organisations have been added to the existing list: Magic Minimax 01.ai NVIDIA The above organisations, in furtherance of safe and trustworthy AI, undertake to develop and deploy their frontier AI models and systems responsibly, in accordance with the following voluntary commitments, and to demonstrate how they have achieved this by publishing a safety framework focused on severe risks by the upcoming AI Summit in France.”