Amazon has introduced the Amazon Nova AI model suite, comprising up to 6 models with varying capabilities.
Amazon Nova Micro: A small model that reads text only, emphasizes quick response, affordable pricing, supports up to 128,000 tokens, with input price at $0.035 per million tokens and output at $0.14 per million tokens.
Amazon Nova Lite: A compact model that supports text, images, videos, similarly focusing on budget-friendly pricing, supports 300,000 tokens, with input price at $0.06 per million tokens and output at $0.24 per million tokens.
Amazon Nova Pro: A high-performance model considering price, speed, and accuracy, supports 300,000 tokens, with input price at $0.8 per million tokens and output at $3.2 per million tokens.
Amazon Nova Premier: The top-performing model used for training other models.
Amazon Nova Canvas: A model that creates images from text, pricing depends on image quality, standard pricing at $0.04 per image and up to $0.08 per image.
Amazon Nova Reel: A model that generates videos, 720p resolution at 24 frames per second, priced at $0.08 per second.
The special feature is the LLM model in the Multimodal group, which supports fine-tuning in all formats – text, images, and videos alike.
The testing scores of Nova Micro, Lite, and Pro align with GPT-4o, Gemimi 1.5 Pro, and Claude 3.5 Sonnet for major tasks like programming, finance, and research. However, the interesting part is the GroundUI-1K testing where Nova Pro and Nova Lite outperform others significantly.
Amazon has announced plans to develop a speech-to-speech model for direct conversation by the first quarter of 2025 and aims to advance towards an any-to-any model, allowing input and output customization by mid-2025.
Source – Press Release, About Amazon, Amazon.Science
TLDR: Amazon introduces the diverse Amazon Nova AI model suite with different capabilities catering to various needs. Nova Pro and Nova Lite excel in specific testing scenarios, setting the stage for future advancements in speech technology.
Leave a Comment