Amazon Nova Micro, Lite and Pro
Amazon launched its first-party Nova understanding models on Bedrock, text-only Micro (128K context) and multimodal Lite and Pro (300K), with fine-tuning and distillation.
Replaces the earlier Titan text line with models Amazon positions on price-performance for RAG and agent workloads. Lite takes up to 30 minutes of video per request; Pro handles 15,000+ line codebases. Premier was announced as still in training.
- Date
- Tuesday, 3 December 2024
- Lab
- Amazon
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Context windows | Micro 128K; Lite and Pro 300K tokens Lite takes up to 30 min of video per request. | company |
Announced at re:Invent 2024. The AWS posts opened give no per-token prices or independent benchmark numbers; Titan-era history was not researched.
Sources
- aws.amazon.com/about-aws/whats-new/2024/12/amazon-nova-foundation-models-bedrock/
- aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leadin
This record was checked against its sources on 6 October 2026. How we check