Nova Micro

Amazon Nova Micro sits at the lightweight end of the Nova family of foundation models served through Amazon Bedrock, joining Nova Lite and Nova Pro as a compact, cost-oriented text option for developers who need conversational AI without the overhead of larger endpoints. The model is designed for high-volume, latency-sensitive workloads where predictable throughput matters more than frontier reasoning, and it has already been deployed alongside Nova Lite in documented Bedrock pipelines for structured business data such as claims processing, where the combination is pitched as a way to balance speed and quality while keeping inference spend low.

Beyond raw inference, Nova Micro is treated as a customization-friendly foundation model: Bedrock now supports both supervised fine-tuning and reinforcement fine-tuning for it, letting teams upload JSONL conversational training data from S3 and pay per training token processed. This makes it a practical base for organizations that want to specialize a small language model on a narrow domain, then deploy that tuned variant through the same managed Bedrock environment used for the rest of the Nova lineup, rather than standing up separate infrastructure.