Nova
Amazon's model family, built into AWS Bedrock.
Olya's profile
- What sets it apart
- Nova isn't a chatbot you open and use: it's a family of models that lives only inside AWS Bedrock, reachable by API. It spans the full range of modalities under one brand — text, images, video, voice (Sonic), image generation (Canvas) and video generation (Reel) — with four understanding sizes, from Micro to Premier, ranging from 128k to 1M context. Since December 2025 there's also Nova 2, the generation with extended reasoning and thinking intensity adjustable across three levels.
- Strengths
- Price is the real argument: Micro starts at $0.035 per million input tokens, among the lowest on Bedrock, and the tiers step up on a gradual logic all the way to Premier. Integration with the AWS ecosystem is native — IAM, Bedrock Agents, Knowledge Bases, Guardrails — and Premier acts as a teacher to distill smaller, specialized models. Nova 2 adds a one-million-token context window, a code interpreter, web grounding and support for remote MCP tools.
- When to use it
- It makes sense if you're already inside AWS and want to stay there: high-volume workloads where cost per token matters more than the last benchmark point, support chatbots, document processing, RAG and agents that live alongside your other Amazon services. The small tiers, Micro and Lite, are ideal for summarization, classification and translation at minimal cost; Premier and Nova 2 for multi-document analysis and more complex agentic tasks.
- When to avoid it
- If you don't use AWS, it isn't worth the candle: there's no app or public interface, everything runs through Bedrock and ties you to Amazon's infrastructure. On pure reasoning benchmarks the flagship models from Anthropic, OpenAI and Google stay ahead, so for absolute quality look elsewhere. And remember that Nova 2 Pro is still in a preview reserved for Nova Forge customers: if you need it in production today, it isn't there.