Gemma 3n / Nova Micro

Gemma 3n (Google) and Nova Micro (Amazon Bedrock) represent the cheapest tier of capable AI models in 2026, priced at approximately $0.02–$0.03 per million tokens. They are designed for edge deployments, on-device inference, and workloads where cost at extreme scale is the primary constraint. Gemma 3n runs efficiently on consumer hardware via tools like Ollama, while Nova Micro is accessible through AWS Bedrock for managed cloud inference at minimal cost.