Applications of Foundation Models
Amazon Nova Model Family
Amazon Nova is Amazon's family of foundation models served through Amazon Bedrock. The understanding models, Amazon Nova Micro, Lite, Pro, and Premier, accept inputs and produce text output, while the creative models, Amazon Nova Canvas, Reel, and Sonic, generate images, short videos, and real-time conversational speech.
Concepts
- Amazon Nova Micro is the only text-only Nova understanding model and is the lowest-cost, lowest-latency option with about a 128K token context window.
- Amazon Nova Lite, Pro, and Premier are multimodal understanding models that accept text, image, and video input and output text.
- Amazon Nova Canvas generates images, Amazon Nova Reel generates short videos, and Amazon Nova Sonic supports real-time speech-to-speech conversations.
- Nova Micro, Lite, Pro, and Premier support more than 200 languages and are optimized for a core set including English, German, Spanish, French, Italian, Japanese, Korean, Arabic, Simplified Chinese, Russian, Hindi, Portuguese, Dutch, Turkish, and Hebrew.
- Context windows increase from about 128K tokens for Micro to about 300K for Lite and Pro, and about 1M for Premier.
- The understanding model ladder is Micro for cheapest and fastest text tasks, then Lite, Pro, and Premier for the most capable and largest-context reasoning tasks.
Exam tips
- Micro is text-only; Lite, Pro, and Premier are the multimodal text-output understanding models.
- Understanding models do not generate images, video, or audio; Canvas, Reel, and Sonic handle those creative modalities.
- Canvas means image generation, Reel means video generation, and Sonic means speech-to-speech.
- All Amazon Nova models are accessed through Amazon Bedrock.